VISVESVARAYA TECHNOLOGICAL UNIVERSITY

BELAGAVI, KARNATAKA -590 018

A Project Report on

“MULTI-CLASS ADAPTIVE ACTIVE LEARNING FOR

PREDICTING STUDENT ANXIETY”

Submitted in partial fulfillment for the award of degree of Bachelor of Engineering in Computer
Science & Engineering during the year 2024-25
By
Amrutha E 4MH21CS003
Harshitha M R 4MH21CS034
Jeevitha S R 4MH21CS038
Madhumitha R 4MH21CS046

Under the Guidance of

Prof. Prasanna G
Assistant Professor
Designation of Guide,
Dept of CS&E,
MIT Mysore.

2024-25
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
MAHARAJA INSTITUTE OF TECHNOLOGY MYSORE
BELAWADI, NAGUVANAHALLY POST, S.R. PATNA TALUK, MANDYA DIST-571477.
 MAHARAJA INSTITUTE OF TECHNOLOGY MYSORE
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

~~~~~~ CERTIFICATE ~~~~~~

Certified that the project work entitled “MULTI- CLASS ANDAPTIVE

LEARNING FOR PREDICTING STUDENT ANXIETY” is a bonafide work carried
out by Amrutha E (4MH21CS003), Harshitha M R (4MH21CS034), Jeevitha S R
(4MH21CS038) and Madhumitha R (4MH21CS046) in the partial fulfillment for the
award of degree of Bachelor of Engineering in Computer Science & Engineering of the
Visvesvaraya Technological University, Belagavi during the academic year 2024-25. It
is certified that all corrections/suggestions indicated have been incorporated in the
report. The project report has been approved as it satisfies the academic requirements
with respect to the Project work prescribed for Bachelor of Engineering Degree.

Signature of the Guide Signature of the HOD Signature of the Principal

Prof. Prasanna G Dr. Shivamurthy R C Dr. Murali S
Assistant Professor Professor & Head Principal
Dept. of CS&E Dept. of CS&E MIT Mysore

External viva

Name of the Examiners Signature with date

1)

2)
~ ~ ~ ~ ~ ~ ~ ~ ACKNOWLEDGEMENT ~ ~ ~ ~ ~ ~ ~ ~

We, the undersigned, would like to express our sincere gratitude to everyone who
supported and guided us throughout the successful completion of our project.

First and foremost, we extend our heartfelt thanks to our project guide, Prof. Prasanna
G, for their valuable guidance, constant encouragement, and continuous support, which played
a crucial role in shaping the direction and outcome of this project.

We are also deeply thankful to Dr. Shivamurthy R C, Head of the Department of

Computer Science and Engineering, for providing us with the required facilities and an
environment conducive to learning and innovation.

Our sincere thanks also go to all the faculty members and staff of the department for their
cooperation, insightful feedback, and technical support throughout the duration of this work.

Last but not least, we would like to express our deepest gratitude to our families and
friends for their unwavering support, patience, and encouragement, which motivated us to work
with commitment and confidence.

Amrutha E 4MH21CS003
Harshitha M R 4MH21CS034
Jeevitha S R 4MH21CS038
Madhumitha R 4MH21CS046
 ~ ~ ~ ~ ~ ~ ~ ~ ABSTRACT ~ ~ ~ ~ ~ ~ ~ ~

This research introduces a multi-class adaptive active learning framework to predict

student anxiety, aiming to enhance early intervention and support mechanisms in educational
environments. Traditional anxiety prediction models often fall short due to limited labeled data
and static learning processes. Our approach leverages adaptive active learning to iteratively
select the most informative data points for labeling, improving model accuracy and robustness.
By incorporating multi-class classification, the model differentiates between various levels of
anxiety, providing a nuanced understanding of student mental health. Experimental results
demonstrate the effectiveness of the proposed method, showing significant improvements in
prediction accuracy over baseline models. This study underscores the potential of adaptive
active learning in educational data mining, offering a scalable solution for real-time anxiety
prediction and contributing to more responsive and supportive educational systems.
Keywords: Adaptive Active Learning, Multi-ClassClassification, Student Anxiety
Prediction, Educational Data Mining,Machine, Learning in Education,Mental ,Health
Assessment,Real-Time Analytics.
 ~ ~ ~ ~ ~ ~ ~ ~ CONTENTS ~ ~ ~ ~ ~ ~ ~ ~

1. INTRODUCTION ………………………………………...……………… 01
1.1 Overview …………………………...…………………………...… 01
1.2 Problem Statement ……………….……………...……………...… 01
1.3 Solution……………….….…………………………………...…… 01
1.4 Existing System ...…….….…………………………………...…… 01
1.5 Proposed System ….….….…………………………………...…… 02

2. LITERATURE SURVEY …………...……………………….…………... 03

2.1 Literature Review …………………..……….…………………….. 03
2.2 Survey Findings …………………….……………………………... 04

3.SOFTWARE REQUIRMENT SPECIFICATIONS ...………………...... 06

3.1 Functional Requirements ………………...……………….....……. 06
3.2 Non-Functional Requirements ………………...…………..…….... 06
3.3 System Requirements ………………...……………………...……. 07

4. SYSTEM ANALYSIS AND DESIGN ………...………………..……….. 09

4.1 System Analysis………………………….........................………... 09
4.2 High Level Design ………………………….........................…….. 11
4.3 Low Level Design ………………………….........................……... 12
4.4 User Interface Design ……………………….........................……. 18

5. IMPLEMENTATION DETAILS ...…………………...………………… 19

5.1 Control Flow …………………………................................……… 19
5.2 Methodology …………………………...….........................……… 20
5.3 Algorithm ………………………………….........................……… 20
5.4 Source Code ………………………………………………………. 22

6. TESTING DETAILS ……………………………………………………... 28

6.1 Unit Testing …..………………………................................……… 28
6.2 Integration Testing …..……………………….....................……… 28
 6.3 User Testing …..………………………...................................…… 29

7. RESULTS DISCUSSION ………………………………………………... 32

7.1 Snapshots ……...………………………................................……… 32
7.2 Result Discussion ……..……………………….....................……… 37

CONCLUSION ……………………………………………………………. 38

FUTURE ENHANCEMENT ……………………………………………... 39

REFERENCES …………………………………………………………….. 41
 ~ ~ ~ ~ ~ ~ ~ ~ List of Figures ~ ~ ~ ~ ~ ~ ~ ~

1. Fig. 4.2 – High-Level Design of the System……………………………………..…23

2. Fig. 4.3.1 – Use Case Diagram………………………………………………..…….25
3. Fig. 4.3.2 – Class Diagram of the System…………………………………………..26
4. Fig. 4.3.3 – Sequence Diagram of the System………………………………………27
5. Fig. 4.3.4 – Collaboration Diagram of the Order Management System………….…28
6. Fig. 4.3.5 – Deployment Diagram of the System……………………………………29
7. Fig. 4.3.6 – Activity Diagram Representing Workflow and Control Flow………….30
8. Fig. 4.3.7 – Component Diagram of the System…………………………...………..31
9. Fig. 4.3.8 – ER Diagram of the Database……………………………………………32
10. Fig. 4.3.9 – Level 1 Data Flow Diagram of the System……………………….…….33
11. Fig. 4.3.10 – Level 2 Data Flow Diagram Showing Detailed System Processes…....34
12. Fig. 5.1 – Control Flow of the Proposed System…………………………………....38
 ~ ~ ~ ~ ~ ~ ~ ~ List of Tables ~ ~ ~ ~ ~ ~ ~ ~

1. Table 2.1.1 – Comparison of Different Learning Approaches……………..06

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CHAPTER – 1

INTRODUCTION
1.1 Overview
The increasing awareness of mental health issues among students has highlighted the need for
effective and timely interventions. Traditional anxiety prediction models face limitations due to the
scarcity of labeled data and the static nature of their learning processes. This inadequacy hinders the
ability to identify students who are at risk and offer them appropriate support. Addressing this gap is
crucial for enhancing educational environments and ensuring that students receive the help they need
before anxiety issues escalate. By leveraging advanced techniques in machine learning, particularly
adaptive active learning, we can significantly improve the prediction accuracy and robustness of
anxiety models. This research is motivated

1.2 Problem Statement

The problem addressed in this research is the inadequacy of traditional anxiety prediction
models in educational environments, primarily due to their reliance on limited labeled data and static
learning processes. These conventional models struggle to accurately identify and differentiate
between varying levels of student anxiety, which hinders effective early intervention and support. The
research seeks to resolve these issues by implementing a multi-class adaptive active learning
framework that dynamically selects informative data points for improved labeling, thereby enhancing
the prediction accuracy and robustness of the anxiety detection system. This adaptive approach allows
for a more nuanced understanding of student mental health, facilitating timely and appropriate
educational support.

1.3 Solution
This project proposes a multi-class adaptive active learning framework for predicting student
anxiety levels with greater accuracy and efficiency. By dynamically selecting the most informative
and uncertain data points for labeling, the model continuously improves its performance with minimal
human annotation. This adaptive learning cycle allows the system to better differentiate between
varying anxiety levels and enables early detection, ensuring timely intervention and support for
students at risk.

1.4 Existing System

Existing anxiety prediction models typically use static machine learning approaches such as
decision trees, SVMs, or logistic regression trained on pre-labeled datasets. These systems rely heavily
on a large volume of labeled data and do not adapt well to new or changing data patterns. Their
performance is often limited by the quality and quantity of the initial training data.

A major drawback of these systems is their inability to handle real-time data or adapt to
evolving student behavior. They fail to prioritize the most relevant or uncertain data, resulting in
inefficient learning and a lack of precision when distinguishing between mild, moderate, or severe
anxiety cases. This limits their effectiveness in early intervention and student support.

Dept. of CS&E, MIT Mysore 2024-2025 1

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

1.5 Proposed System

The proposed system integrates adaptive active learning with multi-class classification to
enhance anxiety prediction accuracy. Instead of randomly selecting data for training, the system
actively queries the most informative instances, thereby reducing the labeling burden. It adapts to new
data over time, making the model more robust and suitable for real-world educational environments.

Compared to the existing static models, this approach provides several advantages: it improves
prediction accuracy even with fewer labeled samples, captures dynamic behavioral changes among
students, and ensures better identification of varying anxiety levels. This makes the model highly
effective for early detection and timely intervention in student mental health care.

1. Active Learning Strategy

• Instead of selecting training data randomly, the model uses active learning to intelligently
query the most informative and uncertain data samples from a pool of unlabeled instances.

• This approach significantly reduces the labeling burden for experts and ensures that the training
process focuses on the most impactful data points.

2. Multi-Class Classification

• The system is designed to classify anxiety into multiple levels (e.g., none, mild, moderate, and
severe), providing a nuanced understanding of students' mental states.

• Multi-class models allow for more precise interventions based on the severity level.

3. Adaptive Learning Capability

• The model continually updates itself as new data becomes available.

• It adapts to changing student behaviors, academic stressors, and environmental conditions over
time.

• This adaptability ensures that the system remains accurate and relevant in dynamic, real-world
educational settings.

Dept. of CS&E, MIT Mysore 2024-2025 2

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CHAPTER – 2
LITERATURE SURVEY
2.1 Literature Review
a) Topic Overview

The central theme explored in this literature review is the application of active learning and adaptive
learning techniques, particularly in multi-class classification tasks related to student performance
and mental health prediction. The reviewed literature spans journal articles, conference papers, and
comparative studies, reflecting diverse approaches and technological advancements in improving
prediction accuracy while minimizing data labeling efforts.

b) Problems and Issues Addressed

The literature identifies several core challenges:

• The high cost of labeling data for supervised learning models, especially in domains like
mental health.

• The limited adaptability of static learning models, which fail to adjust to evolving data
patterns.

• Difficulty in early identification of students at risk of mental health issues or poor academic
performance.

• Inefficiencies in traditional sampling methods, which may not prioritize informative data
points.

c) Technical Approaches: Existing Solutions vs. Proposed Solutions

Existing Solutions

Traditional models often rely on static datasets and conventional machine learning algorithms such as
decision trees, SVMs, and logistic regression. These systems:

• Use uniform sampling or random data selection, leading to sub-optimal training efficiency.

• Are less dynamic, unable to adjust in real-time to new data trends.

• Require large amounts of labeled data for accurate prediction.

Proposed Solutions (from reviewed sources)

1. Active Learning for Multi-Class Classification (Zhou & Zhang, 2015): Proposes intelligent
sample selection strategies like uncertainty sampling and query-by-committee to reduce
labeling cost and improve accuracy in multi-class settings.

Dept. of CS&E, MIT Mysore 2024-2025 3

 Multi- Class Adaptive Active Learning for Predicting Student Anxiety

2. Adaptive Learning for Educational Data (Kumar & Gupta, 2018): Introduces models that
dynamically adjust based on incoming data. These systems are better suited for long-term
monitoring of student performance and mental health.

3. Hybrid Models using Active Learning (Singh & Choi, 2020): Combines multi-class
classification with active learning to fine-tune mental health prediction. This iterative model
improves differentiation across mental health categories.

4. Adaptive Active Learning for Early Intervention (Miller & Smith, 2019): Focuses on real-
time identification of at-risk students. It applies adaptive learning to prioritize data relevant to
anxiety and performance.

5. Dynamic Sampling Strategies (Wang & Liu, 2021): Compares multiple dynamic sampling
techniques within active learning frameworks. It shows that intelligently chosen data samples
significantly improve prediction outcomes.

Approach Benefits Issues

Easy to implement, requires fewer

Static Learning Poor adaptability, high labeling cost
computational resources

Initial model dependency, may require

Active Learning Efficient labeling, higher model accuracy
expert intervention

Real-time adaptability, handles dynamic Complexity in implementation, may

Adaptive Learning
data overfit to recent trends

Hybrid(Active+ Combines strengths of both models, ideal Higher computational cost, needs well-
Adaptive) for critical prediction tasks tuned models

Tabel 2.1: Comparison of Different Learning Approaches.

2.2 Survey Findings

Key Findings

• Active and adaptive learning techniques significantly improve performance in multi-class

classification tasks, particularly for anxiety and mental health prediction.

• Dynamic sampling strategies allow better model training by focusing only on the most
informative and uncertain data points.

• Adaptive systems offer promising avenues for real-time support in education and healthcare,
particularly for early detection and intervention.

Dept. of CS&E, MIT Mysore 2024-2025 4

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Overall Impression and Assessment

The reviewed systems collectively demonstrate that combining active learning and adaptive
techniques is a powerful approach to solving problems involving high-dimensional, dynamic data.
Especially in sensitive domains like student mental health, these systems show great potential in
minimizing labeling effort, enhancing model precision, and supporting early, data-driven decisions.
However, challenges remain in balancing model complexity, interpretability, and real-world
scalability.

The overall impression of the reviewed systems suggests a promising evolution in the field of
intelligent data-driven prediction, particularly in sensitive and dynamic domains like student mental
health. By integrating active learning with adaptive mechanisms, these systems effectively address
one of the most pressing challenges in machine learning—acquiring high-quality labeled data in a
resource-efficient manner. Active learning ensures that only the most informative data points are
selected for labeling, significantly reducing human effort while maintaining or even enhancing the
model’s performance. Meanwhile, adaptivity allows the models to evolve in real-time, responding to
shifts in student behavior, academic pressure, and other contextual variables. This dynamic response
capability is particularly valuable in mental health monitoring, where student conditions can change
rapidly and unpredictably.

Moreover, these systems promote early detection and timely intervention, which are critical for mental
health support. Through continuous learning and re-training, they can capture complex, high-
dimensional patterns that static models often miss. However, despite these strengths, practical
implementation still faces several hurdles. One key concern is model complexity—highly adaptive
systems often rely on sophisticated algorithms that can be difficult to interpret, posing a challenge for
stakeholders like counselors or educators who may not have technical expertise. Additionally,
ensuring scalability without compromising accuracy and ethical standards remains a significant
concern. Privacy, data security, and system transparency must also be addressed for such models to
gain widespread acceptance and integration into educational infrastructures. Overall, while the
trajectory is promising, future developments must focus on balancing accuracy, interpretability, and
practicality to realize the full potential of adaptive, active learning systems in real-world settings.

Dept. of CS&E, MIT Mysore 2024-2025 5

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CHAPTER – 3

SOFTWARE REQUIRMENT SPECIFICATIONS

3.1 Functional Requirements

1. DATA labeling and selection

• The system must be able to iteratively select the most informative data points for labeling.
• The framework should support the integration of new labeled data to continuously improve
the model.

2. Multi-class classification

• The model must classify anxiety levels into multiple classes (e.g., low, medium, high).
• It should provide predictions that distinguish between different levels of student anxiety.

3. Model accuracy improvement

• The system must enhance the accuracy of anxiety predictions compared to baseline
models.
• It should adapt and update its learning process based on newly labeled data to refine
predictions.

4. Real-time prediction

• The system should be capable of real-time anxiety prediction to enable timely intervention.

5. Data integration

• The framework must integrate with educational data sources to obtain relevant student
information for prediction.

6. User interface

• Provide an interface for educators or administrators to view and interpret anxiety

predictions and levels.
• Allow users to input and label new data points.

3.2 Non-Functional Requirements

1. Scalability

• The system should handle increasing amounts of data and users without performance
degradation.

Dept. of CS&E, MIT Mysore 2024-2025 6

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

• It should be scalable to accommodate different educational institutions and varying data

volumes.

2. Performance

• The framework must provide predictions with minimal latency to support real-time
applications.
• The system should be efficient in processing and selecting data points for labeling.

3. Accuracy and reliability

• Ensure high prediction accuracy and reliability of the anxiety levels.

• The model should be robust against variations in data quality and distribution.

4. Security and privacy

• Protect sensitive student data and ensure compliance with relevant data protection regulations
(e.g., gdpr, ferpa).
• Implement secure data storage and transmission protocols.

5. Usability

• The user interface should be intuitive and easy to use for educators and administrators.
• Provide clear and actionable insights based on the predictions.

6. Adaptability

• The system should adapt to changes in the educational environment or student population.
• It should allow for updates and enhancements to the model and framework based on user
feedback and new research.

7. maintainability

• The system should be designed for easy maintenance and updates.

• Provide documentation and support for troubleshooting and system upgrades.

3.3 System Requirements

3.3.1 Hardware Requirements:

Processor - I3/Intel Processor

Hard Disk - 160GB
Key Board - Standard Windows Keyboard
Mouse - Two or Three Button Mouse

Dept. of CS&E, MIT Mysore 2024-2025 7

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Monitor - SVGA
RAM - 8GB

3.3.2 Software Requirements:

• Operating System : Windows 7/8/10

• Server side Script : HTML, CSS, Bootstrap & JS
• Programming Language : Python
• Libraries : Flask, Pandas, Mysql.connector, Os, Scikit-learn, Numpy
• IDE/Workbench : PyCharm
• Technology : Python 3.6+
• Server Deployment : Xampp ServeR

The successful implementation of the anxiety prediction system requires a balanced configuration
of both hardware and software components. On the hardware side, a machine equipped with an Intel
Core i3 processor or above is recommended to ensure reliable performance during code execution,
model training, and server deployment tasks. The system should have at least 160GB of storage space
to accommodate the operating system, project files, dependencies, and datasets. An SVGA or higher-
resolution monitor is necessary for clear display of visual outputs, while a standard Windows-
compatible keyboard and a two- or three-button mouse will support effective user interaction during
development and testing phases. Additionally, a minimum of 8GB RAM is essential for smooth
multitasking, especially when handling large datasets or executing memory-intensive Python
operations, ensuring that tools like Flask, Scikit-learn, and Pandas function without performance
bottlenecks.

On the software front, the system is built to run on Windows operating systems such as Windows
7, 8, or 10, providing compatibility with a wide range of development tools. The user interface of the
web application is designed using HTML, CSS, Bootstrap, and JavaScript, ensuring a responsive and
user-friendly experience. Python serves as the core programming language due to its robust support
for machine learning and data analysis. Key Python libraries utilized include Flask for web application
development, Pandas and NumPy for data handling, Scikit-learn for implementing machine learning
algorithms, and mysql.connector for database interaction. The Integrated Development Environment
(IDE) used is PyCharm, which offers powerful code editing, debugging, and project management
features. For backend support, XAMPP is used to simulate a local server environment, making it easy
to test and manage the MySQL database. The system is developed using Python 3.6 or higher, ensuring
compatibility with the latest library versions and features. This software configuration creates a stable,
scalable, and efficient development environment suitable for building and deploying an intelligent,
real-time student anxiety prediction system.

Dept. of CS&E, MIT Mysore 2024-2025 8

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CHAPTER – 4

SYSTEM ANALYSIS AND DESIGN

4.1 System Analysis
4.1.1 Existing System

The existing system for predicting student anxiety utilizes a variety of machine learning
algorithms, each offering unique strengths in handling the complexities of mental health
assessment within educational environments:

K-Nearest Neighbors (KNN): In the current system, the KNN algorithm is implemented to
classify levels of anxiety based on similarity measures. It operates on the principle that students
with similar behavioral and emotional features likely experience similar levels of anxiety. This
method is particularly useful for capturing local patterns in the data, but its performance can
degrade with high-dimensional data typically found in student behavioral assessments.

Logistic Regression (LR): Logistic Regression is employed to provide a probabilistic foundation

for anxiety prediction, modeling the probabilities of different levels of anxiety as a function of
student characteristics and behaviors. This algorithm is straightforward and effective for binary
classification problems but extends to multi-class problems to estimate the probability of each
category of anxiety, providing a baseline for performance comparisons with other algorithms.

XGBoost (XGB): XGBoost, an implementation of gradient boosted decision trees designed for
speed and performance, is particularly adept at handling varied data types and complex structures
that are common in educational data. It boosts the model's performance by focusing on correcting
the predecessor's errors, thus being highly efficient in capturing non-linear interactions and
relationships among features.

Naive Bayes (NB): The Naive Bayes classifier is used for its ability to handle large datasets
efficiently. Assuming independence between predictors, Naive Bayes calculates the probability of
certain anxiety levels given a set of observed features. It's particularly favored for its simplicity
and speed in making predictions, although its assumption of feature independence may not always
hold in complex educational settings.

Random Forest (RF): Random Forest aggregates multiple decision trees to improve the
classification accuracy and control over-fitting, making it robust against noise and capable of
handling imbalanced data. In the existing system, RF is used to evaluate its effectiveness across a
spectrum of anxiety classifications by leveraging its ensemble learning technique, which enhances
generalizability and accuracy in diverse educational datasets Each of these algorithms contributes
differently to the overall prediction system, leveraging their respective strengths to enhance the
accuracy and reliability of anxiety predictions in students. This multiplicity of approaches helps to
ensure robustness and adaptability in the face of varied data characteristics and evolving
educational environments.

Dept. of CS&E, MIT Mysore 2024-2025 9

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

4.1.2 Disadvantages

1. Complexity in Implementation: Adaptive active learning systems are inherently complex due to
the need for continuous model retraining and data reevaluation. This complexity could pose challenges
in practical educational settings where resources and technical expertise might be limited.

2. Dependence on Initial Data: The success of the adaptive active learning model hinges on the
initial set of labeled data. If this initial data is not representative or sufficiently diverse, the model may
develop biases or fail to generalize well across different student populations.

3. Computational Cost: Continuously selecting informative data points and retraining the model can
be computationally expensive and time-consuming, which might not be feasible in real-time
applications without substantial computational resources.

4. Privacy Concerns: The collection and analysis of sensitive student data such as mental health
indicators require stringent data privacy measures. Ensuring privacy and securing data can be
challenging and increase the operational complexity and costs.

5. Adaptivity to Diverse Educational Settings: While the framework is scalable, its adaptability to
diverse educational environments with varying levels of technology integration and pedagogical
approaches can be limited.

4.1.3 Advantages

1. Enhanced Accuracy: Utilizing decision trees as base learners allows the system to capture intricate
patterns and decision rules from a range of educational data. This, combined with the refinement of
predictions through a stacking classifier, leads to improved accuracy in identifying different levels of
anxiety.

2. Dynamic Learning: The adaptive active learning component of the system dynamically selects
the most informative unlabeled data points for labeling. This process ensures the model continuously
learns and adapts to new data, enhancing its relevance and accuracy over time.

3. Real-Time Analytics: The framework is designed to support real-time analytics, enabling

immediate assessment and intervention. This is crucial for educational settings where timely response
can significantly impact students' well-being and academic performance.

4. Scalability: The system's scalable design allows it to be implemented in various educational

settings, regardless of size, ranging from small classrooms to large institutions. This ensures that it can
benefit a wide demographic of students.

Dept. of CS&E, MIT Mysore 2024-2025 10

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

5. Comprehensive Data Integration: By integrating diverse data sources, such as student

performance, behavioral indicators, and psychometric assessments, the system provides a holistic view
of student anxiety. This comprehensive approach aids in making more informed and nuanced
predictions.

4.2 High Level Design

1. Data Collection and Preprocessing

• Data Sources: Collect data from various sources such as student surveys, academic performance,
attendance records, and psychological assessments.
• Preprocessing: Clean and preprocess the data to handle missing values, normalize features, and encode
categorical variables. Perform feature extraction and selection to identify relevant attributes for anxiety
prediction.

2. Initial Model Training

• Model Selection: Start with baseline models for multi-class classification, such as Logistic Regression,
Decision Trees, or Support Vector Machines (SVMs).
• Training: Train the initial models on the preprocessed data using a standard dataset split (e.g., 70%
training, 30% validation).

3. Adaptive Active Learning Framework

• Data Pooling: Maintain a pool of unlabeled data that can be used for active learning.
• Uncertainty Sampling: Implement uncertainty sampling techniques (e.g., uncertainty sampling,
query-by-committee) to identify data points where the model is least confident.
• Query Selection: Use an adaptive algorithm to iteratively select the most informative samples from
the pool for labeling.
• Labeling Process: Incorporate a feedback mechanism where selected samples are labeled by experts
(e.g., psychologists, educators).

4. Model Iteration and Improvement

• Model Update: Retrain the model with the newly labeled data to improve its accuracy and robustness.
• Evaluation: Continuously evaluate model performance using metrics such as accuracy, precision,
recall, and F1-score for multi-class classification.

5. Anxiety Level Classification

• Multi-Class Classification: Implement the multi-class classification model to predict various levels
of anxiety (e.g., low, moderate, high).
• Output Interpretation: Provide a detailed report on predicted anxiety levels and their corresponding
probabilities.

6. User Interface and Integration

Dept. of CS&E, MIT Mysore 2024-2025 11

Multi- Class Adaptive Active Learning for Predicting Student Anxiety
• Dashboard: Develop a user-friendly dashboard for educators and mental health professionals to view
predictions and insights.
• Integration: Integrate the system with existing educational platforms or systems for seamless access
to predictions and data.

7. Real-Time Monitoring and Feedback

• Monitoring: Implement real-time monitoring to track model performance and user interactions.
• Feedback Loop: Establish a feedback loop to refine and improve the model based on user input and
new data.

8. Scalability and Maintenance

• Scalability: Ensure the system can handle increasing volumes of data and users.
• Maintenance: Regularly update the model and framework.

Fig.4.2.1: High Level design

4.3 Low Level Design

UML Diagrams
UML stands for Unified Modelling Language. UML is a standardized general-purpose
modelling language in the field of object-oriented software engineering. The standard is managed, and
was created by, the Object Management Group.

Dept. of CS&E, MIT Mysore 2024-2025 12

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

The goal is for UML to become a common language for creating models of object-oriented
computer software. In its current form UML is comprised of two major components: a Meta-model
and a notation. In the future, some form of method or process may also be added to; or associated with,
UML.
The Unified Modelling Language is a standard language for specifying, Visualization,
Constructing and documenting the artefacts of software system, as well as for business modelling and
other non-software systems.
The UML represents a collection of best engineering practices that have proven successful in
the modelling of large and complex systems.
The UML is a very important part of developing objects-oriented software and the software
development process. The UML uses mostly graphical notations to express the design of software
projects.

4.3.1 Use Case Diagram

• A use case diagram in the Unified Modeling Language (UML) is a type of behavioral diagram
defined by and created from a Use-case analysis.
• Its purpose is to present a graphical overview of the functionality provided by a system in terms
of actors, their goals (represented as use cases), and any dependencies between those use cases.
• The main purpose of a use case diagram is to show what system functions are performed for
which actor. Roles of the actors in the system can be depicted.

Fig 4.3.1: Use Case Diagram

Dept. of CS&E, MIT Mysore 2024-2025 13

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

4.3.2 Class Diagram

In software engineering, a class diagram in the Unified Modelling Language (UML) is a type of static
structure diagram that describes the structure of a system by showing the system's classes, their
attributes, operations (or methods), and the relationships among the classes. It explains which class
contains information.

Fig 4.3.2: Class Diagram of the System

4.3.3 Sequence Diagram

• A sequence diagram in Unified Modeling Language (UML) is a kind of interaction diagram

that shows how processes operate with one another and in what order.
• It is a construct of a Message Sequence Chart. Sequence diagrams are sometimes called event
diagrams, event scenarios, and timing diagrams

Fig 4.3.3: Sequence Diagram of the System

Dept. of CS&E, MIT Mysore 2024-2025 14

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

4.3.4 Collaboration Diagram

In collaboration diagram the method call sequence is indicated by some numbering technique as shown
below. The number indicates how the methods are called one after another. We have taken the same
order management system to describe the collaboration diagram. The method calls are similar to that
of a sequence diagram. But the difference is that the sequence diagram does not describe the object
organization whereas the collaboration diagram shows the object organization.

Fig 4.3.4: Collaboration Diagram of the Order Management System

4.3.5 Deployment Diagram

Deployment diagram represents the deployment view of a system. It is related to the component
diagram. Because the components are deployed using the deployment diagrams. A deployment
diagram consists of nodes. Nodes are nothing but physical hardware’s used to deploy the application.

Fig 4.3.5: Deployment Diagram of the System

4.3.6 Activity Diagram

Activity diagrams are graphical representations of workflows of stepwise activities and actions with
support for choice, iteration and concurrency. In the Unified Modelling Language, activity diagrams
can be used to describe the business and operational step-by-step workflows of components in a
system. An activity diagram shows the overall flow of control.

Dept. of CS&E, MIT Mysore 2024-2025 15

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Fig 4.3.6: Activity Diagram Representing Workflow and Control Flow

4.3.7 Component Diagram

A component diagram, also known as a UML component diagram, describes the organization and
wiring of the physical components in a system. Component diagrams are often drawn to help model
implementation details and double-check that every aspect of the system's required function is covered
by planned development.

Fig 4.3.7: Component Diagram of the System

4.3.8 ER Diagram

An Entity–relationship model (ER model) describes the structure of a database with the help of a
diagram, which is known as Entity Relationship Diagram (ER Diagram). An ER model is a design or

Dept. of CS&E, MIT Mysore 2024-2025 16

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

blueprint of a database that can later be implemented as a database. The main components of E-R
model are: entity set and relationship set.

An ER diagram shows the relationship among entity sets. An entity set is a group of similar entities
and these entities can have attributes. In terms of DBMS, an entity is a table or attribute of a table in
database, so by showing relationship among tables and their attributes, ER diagram shows the
complete logical structure of a database. Let’s have a look at a simple ER diagram to understand this
concept.

Fig 4.3.8: Entity-Relationship (ER) Diagram of the Database

DFD Diagram

Level 1 Diagram

Fig 4.3.9: Level 1 Data Flow Diagram of the System

Dept. of CS&E, MIT Mysore 2024-2025 17

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Level 2 Diagram:

Fig 4.3.10: DFD Level 2 Showing Detailed System Processes and Data Flow

A Data Flow Diagram (DFD) is a traditional way to visualize the information flows within a system.
A neat and clear DFD can depict a good amount of the system requirements graphically. It can be
manual, automated, or a combination of both. It shows how information enters and leaves the system,
what changes the information and where information is stored. The purpose of a DFD is to show the
scope and boundaries of a system as a whole. It may be used as a communications tool between a
systems analyst and any person who plays a part in the system that acts as the starting point for
redesigning a system.

4.4 User Interface Design

The user interface (UI) of the system is designed to be intuitive, user-friendly, and responsive,
ensuring a smooth and effective interaction between the user and the application. The UI follows a
clean layout with clearly labeled fields, buttons, and result displays to guide the user through the
system's functionality without confusion.

Interactive elements like text fields, dropdown menus, radio buttons, and action buttons (e.g.,
"Submit") are used to gather inputs. Upon submitting the input, the system processes the data and
displays the results (such as predicted output or classification) in a clear, formatted section of the page.
Error messages and input validation feedback are also provided to ensure correct data entry.

The design is responsive, allowing it to work seamlessly across various devices, including desktops,
tablets, and smartphones. Technologies like HTML5, CSS3, and JavaScript are used to build the
frontend, while Bootstrap may be used to enhance responsiveness and aesthetic appeal.

Dept. of CS&E, MIT Mysore 2024-2025 18

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CHAPTER – 5

IMPLEMENTATION DETAILS
5.1 Control Flow
The proposed system employs a novel framework that integrates a multi-class adaptive active learning
strategy with decision tree and stacking classifier methodologies to enhance the prediction of student
anxiety levels. Utilizing decision trees as the base learners, the system first captures the underlying
patterns and decision rules from diverse educational data, such as student performance, behavioral
indicators, and psychometric assessments. These decision trees provide a preliminary classification of
anxiety levels, which are then fed into a stacking classifier. The stacking classifier, comprising an
ensemble of various machine learning models, refines these predictions by learning from the decision
outputs of the decision trees, thus improving the overall prediction accuracy and robustness. This
adaptive active learning component dynamically selects and queries the most informative unlabeled
data points for subsequent labeling, ensuring continuous model improvement and adaptation to new
data. The proposed system is designed to be scalable, enabling real-time analytics in educational
settings to facilitate timely interventions and support for students experiencing anxiety.

Fig 5.1: Control Flow of the Proposed System

Dept. of CS&E, MIT Mysore 2024-2025 19

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

5.2 Methodology

The methodology adopted for this project is based on a structured machine learning pipeline aimed at
predicting sleep disorders from input data. The steps taken to fulfill the project objectives are described
below:

Objective 1: Collect and Preprocess Data

• Method/Procedure: Dataset was sourced from reliable medical records or public datasets.
Preprocessing included handling missing values, encoding categorical features, and scaling
numerical values to improve model performance.

Objective 2: Feature Selection and Engineering

• Method/Procedure: Feature importance techniques were applied (e.g., correlation analysis,

statistical significance tests) to identify key predictors like age, BMI, snoring level, etc.

Objective 3: Model Selection and Training

• Method/Procedure: Various models such as Logistic Regression, Decision Trees, Random

Forest, and SVM were trained. The model showing the best accuracy, precision, and recall was
selected for final deployment.

Objective 4: Model Evaluation

• Method/Procedure: Evaluation metrics like Accuracy, Precision, Recall, F1-Score, and

Confusion Matrix were used to assess the performance of each model.

Objective 5: Deployment and Prediction Interface

• Method/Procedure: The final model was deployed using a web-based interface (e.g., Flask or
Django) to allow users to input symptoms and receive predictions in real-time.

5.3 Algorithm
5.3.1. K-Nearest Neighbors

K-Nearest Neighbors (KNN) is a straightforward yet effective classification algorithm used in various
machine learning tasks. In essence, KNN classifies data points based on the majority class among its
'k' closest neighbors in the feature space. The algorithm computes the distance between the query point
and all other points in the training dataset, typically using metrics like Euclidean or Manhattan
distance. It then identifies the 'k' nearest neighbors and assigns the most common class label among
these neighbors to the query point. KNN is particularly advantageous for its simplicity and ability to
adapt to various types of data without assuming an explicit underlying distribution. In the context of
predicting student anxiety, KNN can effectively categorize anxiety levels by leveraging labeled
examples and providing predictions based on similarity, thus enhancing the model’s ability to capture
nuanced anxiety patterns and improve overall classification accuracy. In your study, KNN achieved

Dept. of CS&E, MIT Mysore 2024-2025 20

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

an accuracy of 0.78, demonstrating its potential to differentiate between various levels of student
anxiety.

5.3.2. Logistic Regression

Logistic Regression is a popular algorithm for binary and multi-class classification problems. In the
context of predicting student anxiety with a multi-class framework, Logistic Regression works by
modeling the probability that a given data point belongs to a particular class. It does this by applying
the logistic function (sigmoid function) to a linear combination of the input features. For multi-class
classification, the model uses a technique called "one-vs-rest" or "softmax" to handle multiple
classes.The logistic function transforms the output into a probability score between 0 and 1, which can
be interpreted as the likelihood of the input data belonging to each class. During training, the algorithm
optimizes the model parameters by minimizing a cost function, typically the cross-entropy loss, which
measures the difference between the predicted probabilities and the actual class labels. The accuracy
of 0.60 indicates that the Logistic Regression model correctly predicted the anxiety levels 60% of the
time, suggesting room for improvement in the model’s performance, potentially through further
tuning, feature engineering, or incorporating more sophisticated algorithms.

5.3.3. XGB Classifier

The XGB Classifier, or Extreme Gradient Boosting Classifier, is a powerful machine learning
algorithm based on gradient boosting techniques. It builds an ensemble of decision trees in a sequential
manner where each tree attempts to correct the errors made by the previous ones. This iterative
approach focuses on improving the model's performance by minimizing the loss function, which
measures the difference between predicted and actual values. XG Boost incorporates regularization
techniques to prevent overfitting and enhance the generalization of the model. In your research project,
this algorithm's robustness and ability to handle complex data patterns contribute to achieving an
accuracy of 0.80 in predicting student anxiety levels. Its adaptive nature makes it well-suited for
handling diverse and dynamic educational data.

5.3.4. Naive Bayes

Naive Bayes is a probabilistic classifier based on Bayes' theorem, assuming that the features used for
classification are conditionally independent given the class label. In the context of predicting student
anxiety levels, the algorithm calculates the probability of a student's anxiety level based on their
features (such as academic performance, social interactions, etc.). It does so by first estimating the
prior probabilities of each anxiety class from the training data. Then, it evaluates the likelihood of each
feature belonging to each class, assuming independence between features. The posterior probability
of each class is computed using Bayes' theorem, and the class with the highest posterior probability is
predicted as the student's anxiety level. Despite its simplicity, Naive Bayes can be effective for multi-
class classification tasks, though in this study, it achieved an accuracy of 0.52, indicating room for
improvement compared to other models.

5.3.5. Random Forest

The Random Forest algorithm operates as an ensemble learning method, constructing multiple
decision trees during training and outputting the mode of the classes (classification) or mean prediction

Dept. of CS&E, MIT Mysore 2024-2025 21

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

(regression) of the individual trees. Each decision tree in the forest is trained on a random subset of
the data and features, which introduces diversity and reduces overfitting. During the prediction phase,
each tree votes for a class, and the class receiving the majority vote is chosen as the final prediction.
This approach allows the model to capture complex patterns and interactions in the data. In your
project, the Random Forest classifier achieved an accuracy of 0.82, demonstrating its effectiveness in
distinguishing between various levels of student anxiety. The high accuracy underscores the model's
capability to leverage the adaptive active learning framework to iteratively improve predictions and
provide valuable insights into student mental health.

5.3.6. Decision Tree Classifier

The Decision Tree Classifier operates as a versatile tool for handling the multi-class classification of
student anxiety levels. The internal mechanism of the Decision Tree Classifier involves recursively
splitting the dataset into subsets based on feature values to create a tree-like model of decisions. At
each node of the tree, the algorithm selects the feature and corresponding threshold that best separates
the data according to a criterion like Gini impurity or information gain. This process continues until
the data in each leaf node belongs to a single class or meets a stopping criterion, such as a maximum
tree depth or minimum sample split. The decision tree structure enables intuitive understanding of how
different features contribute to predictions, and its adaptability in handling various anxiety levels
makes it a fitting choice for predicting and interpreting complex patterns in student anxiety data.

5.3.7. Stacking Classifier

The Stacking Classifier is a powerful ensemble learning technique used in this research to predict
student anxiety levels. It operates by combining the predictions from multiple base models (often
referred to as level-0 models) to improve overall performance. In this approach, the base models
generate predictions based on the input data, and these predictions are then used as input features for
a meta-model (level-1 model), which synthesizes the information to produce the final output. This
method leverages the strengths of various classifiers to enhance predictive accuracy. In our project,
the Stacking Classifier achieved an impressive accuracy of 0.86, demonstrating its effectiveness in
capturing the complex patterns associated with student anxiety. By integrating multiple models'
insights, the Stacking Classifier provides a robust solution that enhances prediction accuracy and
reliability compared to individual models, aligning with the study's goal of improving early
intervention in educational settings.

5.4 Source Code

App.py
from flask import Flask,render_template,redirect,url_for,request

from sklearn.ensemble import StackingClassifier

import numpy as np

import joblib

Dept. of CS&E, MIT Mysore 2024-2025 22

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

app = Flask(__name__)

import joblib

import numpy as np

import mysql.connector

mydb = mysql.connector.connect(

host='localhost',

port=3306,

user='root',

passwd='',

database='Anxiety'

mycur = mydb.cursor()

@app.route('/')

def index():

return render_template('index.html')

@app.route('/about')

def about():

return render_template('about.html')

@app.route('/registration',methods=['POST','GET'])

def registration():

if request.method == 'POST':

name = request.form['name']

email = request.form['email']

password = request.form['password']

confirmpassword = request.form['confirmpassword']

if password == confirmpassword:

# Check if user already exists

sql = 'SELECT * FROM users WHERE email = %s'

Dept. of CS&E, MIT Mysore 2024-2025 23

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

val = (email,)

mycur.execute(sql, val)

data = mycur.fetchone() # Use fetchone to check if user exists

if data is not None:

msg = 'User already registered!'

return render_template('registration.html', msg=msg)

else:

# Insert new user

sql = 'INSERT INTO users (name, email, password) VALUES (%s, %s, %s)'

val = (name, email, password)

mycur.execute(sql, val)

mydb.commit()

msg = 'User registered successfully!'

return render_template('login.html', msg=msg)

else:

msg = 'Passwords do not match!'

return render_template('registration.html', msg=msg)

return render_template('registration.html')

@app.route('/login',methods=['POST','GET'])

def login():

if request.method == 'POST':

email = request.form['email']

password = request.form['password']

sql = 'SELECT * FROM users WHERE email=%s'

val = (email,)

mycur.execute(sql, val)

data = mycur.fetchall()

Dept. of CS&E, MIT Mysore 2024-2025 24

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

if data:

if password == data[0][2]:

return render_template('prediction.html')

else:

msg = 'Password does not match!'

return render_template('login.html', msg=msg)

else:

msg = 'User with this email does not exist. Please register.'

return render_template('login.html', msg=msg)

else:

return render_template('login.html')

@app.route('/prediction', methods=['POST', 'GET'])

def prediction():

if request.method == 'POST':

# Retrieving form data

GAD1 = int(request.form['GAD1'])

GAD2 = int(request.form['GAD2'])

GAD3 = int(request.form['GAD3'])

GAD4 = int(request.form['GAD4'])

GAD5 = int(request.form['GAD5'])

GAD6 = int(request.form['GAD6'])

GAD7 = int(request.form['GAD7'])

SWL1 = int(request.form['SWL1'])

SWL2 = int(request.form['SWL2'])

SWL3 = int(request.form['SWL3'])

SWL4 = int(request.form['SWL4'])

SWL5 = int(request.form['SWL5'])

Game = int(request.form['Game'])

Dept. of CS&E, MIT Mysore 2024-2025 25

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Hours = float(request.form['Hours'])

Gender = int(request.form['Gender'])

Age = int(request.form['Age'])

Work = int(request.form['Work'])

Degree = int(request.form['Degree'])

GAD_T = int(request.form['GAD_T'])

SWL_T = int(request.form['SWL_T'])

# Create feature array for prediction

abc = [[GAD1, GAD2, GAD3, GAD4, GAD5, GAD6, GAD7, SWL1, SWL2, SWL3, SWL4,
SWL5, Game, Hours, Gender, Age, Work, Degree, GAD_T, SWL_T]]

abc_array = np.array(abc)

# Load the trained model

loaded_model = joblib.load('stacking_model.pkl')

# Predict the result using the model

result = loaded_model.predict(abc_array)

# Define message and suggestions based on result

if result == 0:

msg = "Extremely difficult"

suggestions = "It seems like you are facing significant challenges. Consider seeking
professional help, talking to a mentor, or breaking down tasks into smaller steps to reduce stress."

elif result == 1:

msg = "Not difficult at all"

suggestions = "You seem to be handling things well! Keep up the good work and maintain a
balanced approach to avoid burnout."

elif result == 2:

msg = "Somewhat difficult"

Dept. of CS&E, MIT Mysore 2024-2025 26

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

suggestions = "You're finding things a bit challenging. It might help to take short breaks,
reassess your workload, or talk to someone you trust for advice."

elif result == 3:

msg = "Very difficult"

suggestions = "You're going through a tough time. Consider reaching out for support from
friends, family, or a counselor, and take things one step at a time."

else:

msg = "Unexpected result"

suggestions = "It appears there was an issue with the prediction. Please try again or contact
support."

# Render the result with message and suggestions

return render_template('prediction.html', msg=msg, suggestions=suggestions)

# If GET request, render the form

return render_template('prediction.html')

@app.route('/logout')

def logout():

return render_template('index.html')

if __name__ == '__main__':

app.run(debug = True)

Dept. of CS&E, MIT Mysore 2024-2025 27

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CHAPTER – 6

TESTING DETAILS
6.1 Unit Testing
Unit testing is the initial phase of software testing where individual units or components of the
software are tested independently to verify that each performs as designed. A "unit" may refer to a
function, method, procedure, or module in the application. The objective is to isolate each part of the
program and ensure that it behaves correctly in terms of logic, inputs, and outputs.

In the context of our project, each function—such as form input validation, prediction logic, data
preprocessing modules, and database operations—was tested independently. We utilized automated
unit testing tools such as PyTest or unittest in Python to run these tests efficiently. For example, in the
prediction module, we verified if the model correctly handles missing values and gives expected
results for known inputs.

Benefits of unit testing include:

• Early detection of bugs and logic errors

• Simpler debugging due to isolated components

• Improved code reusability and modularity

• Serving as documentation for expected behavior

Unit testing helped ensure a strong foundation by validating each block of the system before
integrating them together.

6.2 Integration Testing

Integration tests are designed to test integrated software components to determine if they
actually run as one program. Testing is event driven and is more concerned with the basic outcome
of screens or fields. Integration tests demonstrate that although the components were individually
satisfaction, as shown by successfully unit testing, the combination of components is correct and
consistent. Integration testing is specifically aimed at exposing the problems that arise from the
combination of components.

Software integration testing is the incremental integration testing of two or more integrated software
components on a single platform to produce failures caused by interface defects.

The task of the integration test is to check that components or software applications, e.g. components
in a software system or – one step up – software applications at the company level – interact without
error.

Dept. of CS&E, MIT Mysore 2024-2025 28

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Test Results: All the test cases mentioned above passed successfully. No defects encountered.

Acceptance Testing

User Acceptance Testing is a critical phase of any project and requires significant participation by the
end user. It also ensures that the system meets the functional requirements.

Test Results: All the test cases mentioned above passed successfully. No defects encountered.

6.3 User Testing

User testing, often referred to as usability testing, is the process of evaluating a system by
observing real users interacting with it. This stage aims to ensure the system is user-friendly,
accessible, and aligns with user expectations.

For our project, user testing involved allowing potential users (students, faculty, or testers) to perform
specific tasks—such as inputting data, generating predictions, or navigating through the system—and
collecting their feedback. We focused on key usability aspects such as:

• Clarity of instructions and labels

• Responsiveness of the user interface

• Accuracy and helpfulness of the outputs

• Speed and performance under typical loads

We used methods like observation, surveys, and direct interviews to gather qualitative and quantitative
feedback. Several iterations of user testing were conducted, each followed by improvements based on
findings.

The primary goals achieved through user testing were:

• Identifying design flaws or confusing interfaces

• Ensuring the layout and color scheme were visually accessible

• Verifying that all core functionalities worked from the user's perspective

• Building user confidence and satisfaction in the system

Ultimately, user testing ensured that our solution was not just technically sound, but also intuitive and
effective for end users.

Functional tests provide systematic demonstrations that functions tested are available as specified by
the business and technical requirements, system documentation, and user manuals.

Functional testing is centered on the following items:

Valid Input : identified classes of valid input must be accepted.

Dept. of CS&E, MIT Mysore 2024-2025 29

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Invalid Input : identified classes of invalid input must be rejected.

Functions : identified functions must be exercised.

Output : identified classes of application outputs must be exercised.

Systems/Procedures: interfacing systems or procedures must be invoked.

Organization and preparation of functional tests is focused on requirements, key functions, or special
test cases. In addition, systematic coverage pertaining to identify Business process flows; data fields,
predefined processes, and successive processes must be considered for testing. Before functional
testing is complete, additional tests are identified and the effective value of current tests is determined.

White Box Testing

White Box Testing is a testing in which in which the software tester has knowledge of the inner
workings, structure and language of the software, or at least its purpose. It is purpose. It is used to test
areas that cannot be reached from a black box level.

Black Box Testing

Black Box Testing is testing the software without any knowledge of the inner workings, structure or
language of the module being tested. Black box tests, as most other kinds of tests, must be written
from a definitive source document, such as specification or requirements document, such as
specification or requirements document. It is a testing in which the software under test is treated, as a
black box .you cannot “see” into it. The test provides inputs and responds to outputs without
considering how the software works.

Test Cases

Registration

Test Case 1: Verify that the registration form accepts valid user inputs and successfully creates a new
account.

Input: Valid username, email, password.

Expected Result: User is successfully registered, and an account is created.

Test Case 2: Verify that the registration form rejects invalid inputs (e.g., weak passwords, invalid
email formats).

Input: Invalid email, weak password.

Expected Result: The system displays appropriate error messages, and registration is not completed.

Test Case 3: Verify that the registration form handles duplicate usernames or emails.

Dept. of CS&E, MIT Mysore 2024-2025 30

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Input: Existing username or email.

Expected Result: The system displays an error message indicating the username or email is already
in use.

Login

Test Case 4: Verify that users can log in with valid credentials.

Input: Valid username and password.

Expected Result: User is successfully logged in and redirected to the dashboard.

Test Case 5: Verify that the login form rejects invalid credentials.

Input: Invalid username or password.

Expected Result: The system displays an error message and does not allow login.

Test Case 6: Verify that the login form handles cases where the user account is inactive or deactivated.

Input: Username and password for an inactive/deactivated account.

Expected Result: The system displays a message indicating the account is inactive.

Dept. of CS&E, MIT Mysore 2024-2025 31

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CHAPTER – 7

RESULTS DISCUSSION
7.1 Snapshots

HomePage: The HomePage serves as the landing page of your application. It provides an overview
of the project's features, objectives, and benefits. Users can navigate to other sections of the application
from this page.

AboutPage: The AboutPage offers detailed information about the project, including its purpose, goals,
and the technology used. It provides background information on the problem being addressed and the
methods employed.

Dept. of CS&E, MIT Mysore 2024-2025 32

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Registration Page: The Registration Page allows new users to create an account with the application.
It typically includes fields for entering personal information such as name, email, password, and
possibly other details like phone number or address. Users need to fill out this form to gain access to
the application's features.

Login Page: The Login Page enables users to access their existing accounts by entering their
credentials. It usually includes fields for entering a username/email and password.

Dept. of CS&E, MIT Mysore 2024-2025 33

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Dept. of CS&E, MIT Mysore 2024-2025 34

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Prediction Page: : The Prediction Page allows users to input data and receive predictions based on
the trained machine learning models. This page typically includes a form or interface for uploading or
entering data (e.g., smartwatch sensor data).

Dept. of CS&E, MIT Mysore 2024-2025 35

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

Outputs

7.2 Result Discussion

7.1 MODULES

System

User

1. System

1.1 Store Dataset

The System stores the dataset given by the user.

1.2 Model Training

This is the process of teaching a machine learning model to make accurate predictions or classifications by
exposing it to a dataset. During this phase, data is prepared and split into training, validation, and test sets. The

Dept. of CS&E, MIT Mysore 2024-2025 36

Multi- Class Adaptive Active Learning for Predicting Student Anxiety
selected algorithm learns from the training data by adjusting its internal parameters to minimize errors in
predictions, using techniques like gradient descent to optimize performance.

1.3 Model Predictions

The system takes the data given by the user and predict the output based on the given data.

2. User

2.Registration

The Registration Page allows new users to create an account by entering their personal information. It includes
fields for username, email, password, and other required details. The page features validation to ensure that all
input data is correct and meets the specified requirements. For example, it checks for valid email formats, strong
passwords, and non-duplicate usernames. Users receive real-time feedback on any errors or issues with their
input, ensuring a smooth and secure registration process.

2.2 Login

Username/Email Field: Checks for valid email formats or existing usernames.

Password Field: Ensures the password meets security requirements (e.g., minimum length, complexity).

Validation Messages: Provides immediate feedback if the input is incorrect or if the account details do not
match.

2.3 Evaluation

User can evaluate the model performance.

Dept. of CS&E, MIT Mysore 2024-2025 37

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

CONCLUSION
Our team has developed a novel multi-class adaptive active learning framework designed to predict
student anxiety more accurately and efficiently. This innovative approach tackles key limitations of
traditional models, which often depend on large amounts of labeled data and lack flexibility in adapting
to changing behaviors. By utilizing adaptive active learning, the system selectively queries the most
informative data points, reducing labeling effort while significantly improving prediction performance
and robustness.

The integration of multi-class classification enables the model to differentiate between varying
levels of anxiety—such as none, mild, moderate, and severe. This finer level of detail is essential for
designing tailored interventions and timely support strategies in educational settings. Through rigorous
testing and evaluation, the framework consistently outperformed baseline models, demonstrating its
effectiveness in capturing dynamic patterns in student behavior.

This work underscores the potential of adaptive learning methods in educational data mining,
providing a scalable and real-time solution for identifying anxiety among students. By enabling more
responsive, data-driven mental health support, our contribution aims to enhance student well-being
and promote healthier academic environments.

Dept. of CS&E, MIT Mysore 2024-2025 38

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

FUTURE ENHANCEMENT
1. Integration of Multi-Modal Data

• Description: Incorporate diverse data sources such as physiological data (e.g., heart rate
variability), behavioral data (e.g., participation in class activities), and academic performance
metrics.
• Benefit: Enhances the robustness of the model by providing a more comprehensive view of
student anxiety.

2. Personalized Intervention Strategies

• Description: Develop personalized intervention recommendations based on the predicted

anxiety levels and individual student profiles.
• Benefit: Tailors support mechanisms to individual needs, improving the effectiveness of
interventions.

3. Real-Time Monitoring and Feedback

• Description: Implement real-time monitoring systems to provide instant feedback and

support to students based on the model’s predictions.
• Benefit: Enables timely interventions and support, potentially reducing anxiety before it
escalates.

4. Dynamic Learning and Adaptation

• Description: Introduce mechanisms for dynamic model adaptation as new data is collected,
allowing the model to continuously improve and stay relevant.
• Benefit: Ensures the model remains accurate and effective over time, even as student
behaviors and anxiety patterns evolve.

5. Incorporation of Ethical and Privacy Considerations

• Description: Develop frameworks for addressing ethical issues and ensuring data privacy in
the collection and use of student data.
• Benefit: Builds trust with stakeholders and ensures compliance with legal and ethical
standards.

6. Enhanced Explainability and Transparency

• Description: Implement techniques to make the model’s predictions and decision-making

processes more transparent and interpretable.
• Benefit: Helps educators and students understand and trust the model’s predictions and
recommendations.

Dept. of CS&E, MIT Mysore 2024-2025 39

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

7. Scalability and Deployment

• Description: Explore methods for scaling the model to larger educational institutions or
systems and integrating it into existing educational tools.
• Benefit: Facilitates broader adoption and use of the framework across diverse educational
settings.

8. Cross-Cultural and Contextual Adaptation

• Description: Adapt the model to account for cultural and contextual differences in student
anxiety and coping mechanisms.
• Benefit: Ensures the model’s predictions and recommendations are relevant and effective
across different cultural contexts.

9. Integration with Student Support Services

• Description: Create pathways for seamless integration with existing student support services
and resources.
• Benefit: Enhances the overall support network available to students and streamlines access to
help.

Dept. of CS&E, MIT Mysore 2024-2025 40

Multi- Class Adaptive Active Learning for Predicting Student Anxiety

REFERENCES
[1] J. Southworth, K. Migliaccio, J. Glover, J. Glover, D. Reed, C. McCarty, J. Brendemuhl, and A.
Thomas, ‘‘Developing a model for AI across the curriculum: Transforming the higher education
landscape via innovation in AI literacy,’’ Comput. Educ., Artif. Intell., vol. 4, Aug. 2023, Art. no.
100127.

[2] X.-Q. Liu, Y.-X. Guo, and Y. Xu, ‘‘Risk factors and digital interventions for anxiety disorders in
college students: Stakeholder perspectives,’’ World J. Clin. Cases, vol. 11, no. 7, pp. 1442–1457, Mar.
2023.

[3] M. K. Khaira, R. L. R. Gopal, S. M. Saini, and Z. M. Isa, ‘‘Interventional strategies to reduce test
anxiety among nursing students: A systematic review,’’ Int. J. Environ. Res. Public Health, vol. 20,
no. 2, p. 1233, Jan. 2023.

[4] K. Meeks, A. S. Peak, and A. Dreihaus, ‘‘Depression, anxiety, and stress among students, faculty,
and staff,’’ J. Amer. College Health, vol. 71, no. 2, pp. 348–354, Feb. 2023.

[5] D. H. Halat, A. Soltani, R. Dalli, L. Alsarraj, and A. Malki, ‘‘Understanding and fostering mental
health and well-being among University Faculty: A narrative review,’’ J. Clin. Med., vol. 12, no. 13,
p. 4425, Jun. 2023.

[6] R. D. Godsil and L. S. Richardson, ‘‘Racial anxiety,’’ Iowa Law Rev., vol. 102, p. 2235, Sep.
2016.

[7] J. Santhosh, D. Dzsotjan, and S. Ishimaru, ‘‘Multimodal assessment of interest levels in reading:
Integrating eye-tracking and physiological sensing,’’ IEEE Access, vol. 11, pp. 93994–94008, 2023.

[8] R. Qasrawi, S. P. Vicuna Polo, D. A. Al-Halawa, S. Hallaq, and Z. Abdeen, ‘‘Assessment and
prediction of depression and anxiety risk factors in schoolchildren: Machine learning techniques
performance analysis,’’ JMIR Formative Res., vol. 6, no. 8, Aug. 2022, Art. no. e32736. [9] P. J. Bota,
C. Wang, A. L. N. Fred, and H. P. Da Silva, ‘‘A review, current challenges, and future possibilities on
emotion recognition using machine learning and physiological signals,’’ IEEE Access, vol. 7, pp.
140990–141020, 2019.

[10] P. Meshram and R. K. Rambola, ‘‘Diagnosis of depression level using multimodal approaches
using deep learning techniques with multiple selective features,’’ Expert Syst., vol. 40, no. 4, p.
e12933, May 2023.

[11] E. Mosqueira-Rey, D. Alonso-Ríos, and A. Baamonde-Lozano, ‘‘Integrating iterative machine

teaching and active learning into the machine learning loop,’’ Proc. Comput. Sci., vol. 192, pp. 553–
562, Jan. 2021.

[12] J. C. Cassady, E. E. Pierson, and J. M. Starling, ‘‘Predicting student depression with measures of
general and academic anxieties,’’ Frontiers Educ., vol. 4, p. 11, Feb. 2019. [13] H. S. Yi and W. Na,
‘‘How are maths-anxious students identified and what are the key predictors of maths anxiety? Insights
gained from Pisa results for Korean adolescents,’’ Asia Pacific J. Educ., vol. 40, no. 2, pp. 247–262,
Apr. 2020.

Dept. of CS&E, MIT Mysore 2024-2025 41