Advances in Machine Learning for Chronic Disease Prediction: A Comprehensive Review

Abstract

Chronic conditions including diabetes, cardiovascular disorders, kidney disease, and Alzheimer's contribute increasingly to the global health burden, highlighting the critical need for early diagnosis and timely intervention to enhance patient care and lower medical expenses. Recent advances in Machine Learning (ML) have demonstrated significant potential in predicting the onset and progression of these conditions by analyzing large-scale medical datasets and uncovering complex patterns often missed by traditional diagnostic methods. These technologies enable faster, more accurate and cost-effective assessments, benefiting both clinicians and patients. This review comprehensively examines the application of ML techniques to the prediction of these four major chronic diseases, highlighting their transformative role in early diagnosis. Deep learning models, such as stacked Artificial Neural Network (stack-ANN) and hybrid Convolutional Neural Network Long Short-Term Memory (CNNLSTM) architectures, have achieved high predictive performance, with reported accuracies reaching up to 99.51% and AUC scores of 1.00 in specific contexts. Boosting algorithms, including XGBoost and LightGBM, also deliver robust results, frequently exceeding 98% of accuracy. The review emphasizes the crucial role of data preprocessing and feature selection in enhancing model interpretability and performance. Ensemble methods, such as bagging and voting classifiers, further contribute to improved predictive outcomes. Despite these advancements, the generalizability of many models remains limited due to heavy reliance on single-source datasets.