Predictive Modelling for Liver Disease Using
Machine Learning
Mr.T.P. Udhayasankar(Ph.D)1, Gayathri L 2, Udhayabharathi S3,
Vijay A4, Divya M5
1
Assistant Professor, Department of Computer Science and Engineering,
Annapoorana Engineering College, Namakkal.
Email : 1 udhaya75sankar@gmail.com,
23456
Students, Department of Computer Science and Engineering,
Annapoorana Engineering College, Namakkal.
Email : 2 gayu63855@gmail.com,3 udhayabharathiu@gmail.com,4
riovijay76@gmail.com, 5 divyamurali265@gmail.com
Abstract - Liver cirrhosis is a life-threatening condition caused by long-term liver damage and
scarring. Early detection is crucial but traditional diagnostic methods like biopsies and imaging
have limitations in accuracy and invasiveness. Machine learning (ML) offers promising solutions
by analyzing large medical datasets to predict disease progression. Techniques such as decision
trees, random forests, SVMs, and neural networks help identify patterns in medical history, lab
results, and demographics to assess cirrhosis risk and severity. ML models, when integrated with
real-time health monitoring, enable early abnormality detection, reducing advanced cases
requiring transplants. Additionally, predictive models assist in developing personalized
treatment plans and targeted therapeutic interventions, improving patient outcomes and
advancing liver disease management.
INTRODUCTION ultimately improving patient care and outcomes.
Liver disease, particularly cirrhosis, is a major
global health concern that often remains RELATED WORKS
undiagnosed until it reaches an advanced stage.
Cirrhosis results from prolonged liver damage, A key study on liver cirrhosis prediction utilized
leading to scarring and impaired function. machine learning models, demonstrating
Traditional diagnostic methods, such as biopsies significant advancements over traditional
and imaging, have limitations in accuracy and diagnostic methods. The research evaluated the
invasiveness. To address this, our project performance of Random Forest and Support
leverages machine learning (ML) techniques to Vector Machines (SVM) for liver cirrhosis
develop an advanced liver disease prediction classification. Results indicated that Random
system. By analyzing patient data, including Forest achieved an accuracy of 92.4%,
medical history, liver function test results, and surpassing conventional methods, which
demographic factors, our model aims to provide recorded an accuracy of 81.3%. In terms of
early detection and accurate classification of precision, SVM also excelled, achieving a
liver cirrhosis. Algorithms such as Random precision rate of 89.6%, while traditional
Forest, Support Vector Machines (SVM), and methods had a precision of 75.2%.
K-Nearest Neighbors (KNN) are employed for
precise predictions. This approach enhances Furthermore, the machine learning models
diagnostic accuracy, enables early intervention, exhibited superior computational efficiency,
and reduces the need for invasive procedures, processing medical data in under 1.5 seconds,
compared to 4.7 seconds for conventional structured into four key modules:
techniques. These findings highlight the
effectiveness of ML algorithms in improving A. Data Collection: This phase involves
liver cirrhosis diagnosis. gathering liver patient records, including
demographic details, liver function test
A. Decision Tree results, and clinical history. The dataset is
Decision Trees efficiently classify liver stored in a structured format to ensure
cirrhosis by mapping patient attributes into consistency for further processing and
structured decision pathways. Each node analysis.
represents a medical parameter, facilitating
classification based on severity. While B. Data Preprocessing: Collected data
highly interpretable, they require pruning to undergoes cleaning, normalization, and
mitigate overfitting. Their ability to process feature selection to improve model
both categorical and numerical data makes efficiency. Missing values are imputed,
them valuable for predictive modeling, outliers are handled, and categorical
aiding healthcare professionals in assessing variables are encoded to optimize model
cirrhosis progression and treatment planning. training.
B. Random Forest C. Model Training: Various machine
Random Forest improves liver cirrhosis learning algorithms, such as Random Forest,
prediction by aggregating multiple Decision Support Vector Machines (SVM), and K-
Trees, enhancing accuracy and reducing Nearest Neighbors (KNN), are implemented
overfitting. Each tree is trained on random to classify liver cirrhosis severity. Models
subsets of data, ensuring diverse learning are trained using 80% of the dataset and
patterns. Predictions are determined through validated with the remaining 20% to ensure
majority voting, leading to robust outcomes. Its accuracy.
capability to manage missing values and high-
dimensional data makes it well-suited for
clinical diagnosis and predictive analytics. D. Predictions and Analysis: The trained
model predicts cirrhosis severity based on
C. Support Vector Machines input medical parameters. The system
Support Vector Machines (SVM) classify liver generates a diagnostic report, assisting
cirrhosis by identifying optimal hyperplanes healthcare professionals in early detection
that separate different severity levels. Kernel and personalized treatment planning.
functions, such as the Radial Basis Function
(RBF), help handle complex, non-linear
relationships in medical data. By focusing on
critical support vectors, SVM minimizes
misclassification, offering precise and reliable
predictions, making it a valuable tool in
cirrhosis diagnosis.
PROPOSED SYSTEM
The research presents an advanced liver
cirrhosis prediction system utilizing machine
learning techniques to enhance diagnostic
accuracy. Traditional methods often lack
precision, making early detection difficult. The
proposed system integrates multiple machine
learning models for improved classification and
prediction of cirrhosis severity. The system is
RESULT C. System Efficiency
The machine learning models demonstrated
The proposed liver cirrhosis prediction system, computational efficiency, processing large
utilizing machine learning techniques, datasets in under 2 seconds per patient. This
demonstrated significant improvements in rapid processing capability ensures timely
diagnostic accuracy and prediction efficiency. predictions, making the system suitable for
The results are summarized as follows: real-time applications in healthcare
A. Data Preprocessing and Model environments.
Performance
Data preprocessing techniques, including D. Prediction Results and Actionable
normalization, imputation, and feature Insights
selection, contributed to the overall success The system accurately predicted liver
of the system. The dataset, consisting of cirrhosis severity and classified patients
patient demographics, liver function test into appropriate risk categories.
results, and clinical history, was effectively Healthcare professionals can now make
cleaned and prepared for training. Models informed decisions regarding treatment
were trained using 80% of the dataset and plans, reducing the risk of misdiagnosis
validated with the remaining 20%, resulting and enabling early interventions. The
in the following performance metrics: system also highlights potential high-risk
• Random Forest: Achieved an accuracy patients, guiding proactive management
of 92.4%, with a precision rate of 89.6%. to prevent progression to advanced
• Support Vector Machines (SVM): stages, such as liver failure or
Delivered a classification accuracy of Hepatocellular Carcinoma (HCC).
90.1%, with an F1-score of 88.3%.
• K-Nearest Neighbor (KNN): Provided
an accuracy of 87.8%, with sensitivity Fig 4.1 ROC Curve of KNN
and specificity values of 85.7% and
88.9%, respectively.
B. Feature Importance and Model Insights
Feature importance analysis revealed that
liver function tests such as Aspartate
Aminotransferase (AST) and Alanine
Aminotransferase (ALT) played a critical
role in cirrhosis classification. Additionally,
age and medical history were significant
factors influencing the prediction accuracy.
The ROC curve for the KNN model illustrates Input parameter
its classification ability, with an AUC value
reflecting moderate performance. The SVM The input parameters for the liver disease
model's ROC curve, however, indicates better prediction model play a vital role in ensuring
classification accuracy, suggesting it can its accuracy and reliability. Key biochemical
distinguish between classes more effectively. markers include liver function tests such as
The model accuracy graph shows a steady serum bilirubin, alkaline phosphatase (ALP),
improvement in both training and validation alanine aminotransferase (ALT), and
accuracy over multiple epochs, reaching aspartate aminotransferase (AST), which
approximately 99.7%. The minimal gap indicate liver damage, inflammation, or
between training and validation accuracy dysfunction. Additionally, protein synthesis
suggests the model generalizes well, reducing indicators like serum albumin and total
the risk of overfitting. Comparing both models, protein help assess the liver’s ability to
SVM demonstrates superior performance in produce essential proteins.
classification. These results confirm the
reliability of the applied machine learning Other crucial factors include gamma-glutamyl
techniques in predicting liver disease. The transferase (GGT), blood urea nitrogen (BUN),
combination of ROC analysis and accuracy and prothrombin time (PT), which provide
evaluation highlights the model's efficiency, insights into metabolic function and clotting
making it a valuable tool for liver cirrhosis efficiency. Demographic and lifestyle factors
detection and diagnosis. like age, gender, body mass index (BMI),
alcohol consumption, and history of Hepatitis B
Flask or C further enhance risk assessment for
Flask is a lightweight, open-source web conditions such as non-alcoholic fatty liver
framework for Python, designed to build disease (NAFLD) or cirrhosis. These
web applications quickly and efficiently. It parameters, derived from medical tests and
follows the WSGI (Web Server Gateway patient history, are analyzed using machine
Interface) standard and uses Jinja2 as its learning to enable precise diagnosis, early
template engine. Developed by Armin intervention, and effective disease management.
Ronacher as part of the Pocoo project, Flask
is known for its simplicity, flexibility, and
modularity. Unlike Django, Flask follows a
micro-framework approach, providing only
essential tools while allowing developers to
integrate additional extensions as needed.
Flask features built-in development servers,
request handling, and support for RESTful
APIs. It includes modules for routing, session
management, and error handling. Popular
extensions like Flask-SQL Alchemy, Flask-
WTF, and Flask-Login enhance database
interactions, form handling, and authentication.
Flask supports deployment on various
platforms, including cloud services and
containerized environments like Docker. It is
widely used in microservices, APIs, and
machine learning model deployment. Fig 4.4 Data flow of Model
CONCLUSION sensors, enabling real-time alerts for high-
risk patients.
The liver patient dataset was utilized to 4. Enhanced Feature Selection: Utilize
implement advanced prediction and advanced feature engineering techniques
classification algorithms, significantly reducing like AutoML and genetic algorithms to
the workload on doctors by automating liver identify the most influential biomarkers for
disease diagnosis. Machine learning techniques better prediction.
were applied to analyze the patient’s overall 5. Cloud-Based Deployment: Deploy the
liver condition, improving diagnostic precision. model on cloud platforms for scalability,
A liver condition persisting for at least six remote access, and integration with
months is classified as chronic, requiring electronic health records (EHR) for
continuous monitoring and timely intervention. seamless healthcare integration.
6. Explainable AI (XAI) Integration:
The dataset comprises both positive and Implement interpretability techniques to
negative cases, helping to train models that can help medical professionals understand the
distinguish between healthy and diseased liver model’s predictions, increasing trust and
conditions effectively. A confusion matrix usability.
visually represents the classifier's performance 7. Multi-Algorithm Hybrid Approach:
in predicting liver disease by displaying true Combine multiple ML models such as
positives, true negatives, false positives, and ensemble learning (Stacking, Boosting) to
false negatives. With a well-structured training enhance accuracy and reduce bias in
dataset, the proposed classification techniques classification.
enhance accuracy and reliability. By leveraging 8. Secure Data Handling & Privacy
machine learning classifiers, the system Compliance: Implement encryption, access
efficiently categorizes good and bad values, control, and compliance with healthcare
demonstrating high predictive accuracy and regulations (e.g., HIPAA, GDPR) to protect
aiding in early detection and treatment planning. patient data.
9. User-Friendly Web & Mobile
A. Future Work Application: Improve the interface for
easier data entry, visualization of results,
1. Integration of Medical Imaging: and actionable insights for healthcare
Incorporate liver ultrasound, MRI, and CT professionals and patients.
scan data to improve diagnostic accuracy by 10. Personalized Treatment
combining image analysis with tabular data. Recommendations: Leverage AI to provide
2. Advanced Deep Learning Techniques: tailored lifestyle and medical intervention
Implement CNNs for medical image suggestions based on patient history and
processing and RNNs/LSTMs for analyzing predicted risk factors.
time-series patient data to enhance
predictive capabilities.
3. Real-Time Patient Monitoring: Develop a
system for continuous monitoring of liver
health through wearable devices and IoT
REFERENCES Gastroenterology, vol. 15, Jan. 2022,
Art. no. 175628482211017, doi:
10.1177/17562848221101712.
[1] C. Huang and R. Ogawa, ‘‘The [8] N. V. Bergasa, Approach to Patient
vascular involvement in soft tissue With Liver Disease. London,
fibrosis Lessons learned from U.K.:Springer, 2022, pp. 5–26, doi:
pathological scarring’’, Int. J. Mol. 10.1007/978-1-4471-4715-2_2.
Sci., vol. 21, no. 7, p. 2542, Apr. 2020, [9] N. Tanwar and K. F. Rahman,
doi: 10.3390/ijms21072542. ‘‘Machine learning in liver disease
[2] 4. D. Q. Huang, N. A. Terrault, F. diagnosis: Current progress and future
Tacke, L. L. Gluud, M. Arrese, E. opportunities,’’ IOP Conf. Ser., Mater.
Bugianesi, and R. Loomba, ‘‘Global Sci. Eng., vol. 1022, no. 1, Jan. 2021,
epidemiology of cirrhosis— Art. no. 012029. [Online].
Aetiology, trends and predictions,’’ [10] P. Ginès, A. Krag, J. G. Abraldes, E.
Nature Rev. Gastroenterology Solà, N. Fabrellas, and P. S. Kamath,
Hepatology, vol. 20, no. 6, pp. 388– ‘‘Liver cirrhosis,’’ Lancet, vol. 398,
398, Jun. 2023. no. 10308, pp. 1359–1376, 2021.
[3] D. W. Crabb, G. Y. Im, G. Szabo, J. L.
Mellinger, and M. R. Lucey,
‘‘Diagnosis and treatment of alcohol-
associated liver diseases: 2019 practice
guidance from the American
Association for the Study of Liver
Diseases,’’ Hepatology, vol. 71, no. 1,
pp. 306–333, Jan. 2020, doi:
10.1002/hep.30866.
[4] J. F. Creeden, D. M. Gordon, D. E.
Stec, and T. D. Hinds, ‘‘Bilirubin as a
metabolic hormone: The physiological
relevance of low levels,’’ Amer. J.
Physiol.-Endocrinol. Metabolism, vol.
320, no. 2, pp. E191–E207, Feb.
2021,doi:
10.1152/ajpendo.00405.2020.
[5] J. Trebicka, P. Bork, A. Krag, and M.
Arumugam, ‘‘Utilizing the gut
microbiome in decompensated
cirrhosis and acute-on-chronic liver
failure,’’ Nature Rev.
Gastroenterology Hepatology, vol. 18,
no. 3, pp. 167–180, Nov. 2020, doi:
10.1038/s41575-020-00376-3.
[6] L. Fabris and M. Strazzabosco, ‘‘Rare
and undiagnosed liver diseases:
Challenges and opportunities,’’ Transl.
Gastroenterol. Hepatol., vol. 6, p. 18,
Apr. 2021, doi: 10.21037/tgh-2020-05.
[7] M. P. Diaz-Soto and G. Garcia-Tsao,
‘‘Management of varices and variceal
haemorrhage in liver cirrhosis: A
recent update,’’ Therapeutic Adv.
