Utilizing Machine Learning Models to
Predict Student Performance
•Name : Wazir Mariyam Hussain
•Reg.No: 2020-KIU-BS2039
•Supervisor : Dr. Muhammad Ismail
•Co-Supervisor: Iqbal Hussain
key points
• Background and Introduction
• Literature Review
• Problem Statement
• Research Objectives
• Study Area
• methodology
• Dataset and Features
• Algorithms Used
• Performance Metrics
• Best Performing Models
• Future work
Background and introduction
• Overview: Importance of education in societal development and how Learning Management
Systems (LMS) produce large datasets. Predicting student performance helps identify students
at risk, allowing educators to improve teaching strategies.
• The role of education in societal development and technological integration.
• Utilization of LMS data to predict and improve student performance.
• Research focus: using machine learning (ML) algorithms to predict student academic
performance.
Literature Review
• Various machine learning techniques such as classification, regression, and
clustering have been applied to predict student performance.
• Studies show that engagement metrics (e.g., login frequency, assignment
submissions, quiz attempts) from LMS data are strong predictors of academic
success.
• Random Forest, Decision Trees, and Artificial Neural Networks (ANN) have
been widely used for prediction, with Random Forest and Decision Trees
showing superior accuracy in several studies.
• Baars & Arnold (2014): ANN successfully predicted student grades based on
LMS activity, with high prediction accuracy.
• The literature indicates that machine learning models are effective in predicting
student performance, with Random Forest and Gradient Boosting often
emerging as top performers.
•LMS platforms generate vast amounts of student data,
yet it is underutilized.
Problem •Need for predictive models that provide early warnings
and insights for educators.
Statement •Goal: use continuous assessment scores and other
features to predict aggregate scores.
Scope and Limitations
Preprocess LMS data to •
Objectives handle missing values and
prepare for machine learning
• Scope: Predict student performance
based on LMS data using machine
(Simple Imputer , learning models.
LabelEncoder).
•
Limitations:
Apply machine learning
models (Neural Networks, Data limited to one
SVM, Decision Tree, etc.) for institution, focuses on
prediction. academic factors only.
Evaluate model
Visualize and interpret
performance using metrics
results for practical insights.
like RMSE and R2.
Study Area
• Institution: Karakorum International University (KIU), Gilgit-Baltistan
• Departments: Focus on specific departments within KIU that utilize LMS for
student management.
• Data Source:
• LMS data logs from KIU, capturing student activities such as:
• Login frequency
• Assignment submissions
• Mid-term and final-term scores
• Sample Size:
•Data from 1,393 students across various courses and departments.
• Study Focus:
• Predicting student performance in different courses using machine learning
models.
• Main academic factors analyzed: Mid-term scores, continuous assessments, and
final exam results.
Methodology
Data Collection: Encoding
Acquired from Handling missing categorical
Karakorum Preprocessing: values with mean attributes like
International imputation. department and
University. subjects.
Model
Data
Development:
Standardization of Transformation:
Applied ML Random Forest,
numerical features Normalizing
algorithms Gradient Boosting.
(e.g., scores). features for model
including SMOreg,
readiness.
MLP,
Dataset and Features
Features in the dataset: Key preprocessing steps:
Academic performance (mid- noise removal, scaling, and
term, continuous assessments, encoding categorical data.
final-term, total marks).
Demographic and academic data
(department, subject, year).
Algorithms Used
Linear Regression: Decision Tree:
Establishes linear Captures nonlinear
relationships between relationships but prone
scores. to overfitting.
Neural Network
(MLP): Models complex Random Forest:
patterns but Ensemble method
computationally offering high accuracy.
intensive.
Gradient Boosting: Gaussian Process:
Iteratively refines weak Probabilistic approach
learners for precise but struggled with
results. larger datasets.
Regression Overview
Definition: Regression is a supervised machine learning
technique used to predict a continuous outcome
(e.g., student performance Performance metrics:
• RMSE
scores) based on input • R2
• PAD
variables. • MSE
Performance metrics
1 2 3
Percentage Absolute Root Mean Square Error R² Score: The R² score
Difference (PAD): PAD (RMSE): RMSE quantifies indicates the proportion of
evaluates prediction prediction accuracy, variance in data explained
accuracy by measuring the reflecting the average by the predictive model,
absolute percentage error deviation of predicted scores assessing its effectiveness.
between predicted and actual from actual student
performance values. outcomes.
Linear Regression Results: Linear Regression
achieved R² of 0.87 and RMSE of 7.37,
indicating strong predictive capability.
Model Results - Decision Tree Performance: The Decision
Tree outperformed with an R² of 0.95 and
Linear Regression RMSE of 4.79, enhancing prediction
& Decision Tree accuracy.
Comparison Insights: Comparative analysis
highlights significant performance
differences, suggesting varying effectiveness
for predicting student outcomes.
Neural Network Performance: The Multi-
Layer Perceptron achieved an R² of 0.85
and RMSE of 7.83, demonstrating effective
prediction.
Model Results - Random Forest Excellence: With an R² of
0.96 and RMSE of 4.27, Random Forest
Linear Regression significantly outperforms other algorithms
& Decision Tree evaluated
Comparative Analysis: Evaluation reveals
Neural Networks deliver solid predictions,
while Random Forest excels in accuracy
and error management.
Gradient Boosting Efficiency:
Achieving R² = 0.96 and RMSE = 4.16,
Gradient Boosting demonstrates
superior predictive performance among
analyzed algorithms.
Model Results - Gaussian Process Limitations:
Despite advanced capabilities, Gaussian
Linear Regression Processes show disappointing
performance characterized by a notably
& Decision Tree high RMSE metric.
Algorithmic Performance
Comparison: The contrasting
performance metrics highlight the
necessity for careful algorithm selection
tailored to educational contexts
Best Performing Models
• Gradient Boosting Effectiveness: High R² and low RMSE indicate Gradient Boosting's robust
capacity to capture complex patterns in educational data.
• Random Forest Advantages: Employing ensemble methods, Random Forest mitigates
overfitting, enhancing model stability and generalization across diverse datasets.
• Interpretable Results: Both models provide interpretable results, aiding stakeholders in
understanding factors influencing student performance outcomes effectively
• Gradient Boosting had the lowest RMSE and highest R², making it the most precise.
•
• Key Findings Recap: The study reveals
Future Work diverse algorithm performances,
underscoring the importance of tailored
predictive modeling for education.
• Future Research Directions: Integrating
demographic factors may enhance model
accuracy by capturing socio-economic
and personal influences on learning.
• Real-Time Interventions: Investigating
real-time interventions could provide
immediate support to students,
improving performance through timely
feedback mechanisms
• I would like to express my heartfelt gratitude to my supervisor, Dr. Ismail and
Mr. Iqbal Hussain, for their unwavering guidance, valuable feedback, and
continuous support throughout this project.
• Their insightful suggestions and encouragement have been instrumental in
shaping the direction and quality of my work.
Thank You
