Enhanced Prediction of Liver Disease Using Machine Learning: A Comparative Analysis

Résumé fourni par la source

Liver disease presents a growing global health burden that often arises slowly without symptoms and hence remains underdiagnosed until late-stage disease, prompting the need for accurate early diagnostic capabilities. With this work, the problem of delayed diagnosis is addressed by using Machine learning (ML) techniques for predicting liver disease from clinical and lifestyle data. In this paper, the principal objective is to assess the accuracy of three existing ML classifiers (RF, LR, and SVM). Using a tidy dataset of 11 features for each of 1700 patients, we went through the usual steps to preprocess and clean our data before splitting it into stratified train-test sets, thereby ensuring an unbiased evaluation. All models were trained/tested using z-scored features and evaluated in terms of classification accuracy, confusion matrices, ROC curves, and statistical analysis. The deep learning models that exceed baseline models have achieved as accuracy (89%, 83%, 85%), sensitivity (87%, 84%, 84%), specificity (92%, 81%, 85%), precision (93%, 84%, 87%), recall (87%, 84%, 84%), F1-Score (90%, 84%, 86%), MSE (0.11,0.17, 0.15), and ROC-AUC (96%, 90%, 92%) for RF, LR, and SVM models, respectively. Random Forest achieved the highest accuracy of 90% and ROC-AUC score of 96%, proving to be very efficient in predicting nonlinear patterns and reducing misclassification errors. The model scored well across multiple metrics, but the weak score in Logistic Regression was due to poor prediction performance. Our results demonstrate the superior performance of ensemble learning models for clinical decision support in the early diagnosis and treatment planning of liver disease. This comparative analysis further supports the use of AI for healthcare diagnostics and serves as a foundation for future multimodal and deep learning approaches.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.