GA-BP Neural Network and Chi-Square Test for CVD, Stroke, and Cirrhosis Comorbidity Risk Prediction

Abstract

Based on big data analytics and machine learning methods, this study develops a series of prediction and prevention models for three major diseases: cardiovascular disease, stroke, and cirrhosis. The datasets underwent systematic preprocessing, including missing value imputation, outlier removal, and categorical variable encoding. Associations between features and diseases were thoroughly examined using chi-square tests and visualization tools such as boxplots and correlation heatmaps, identifying significant factors such as smoking status, ST segment slope, and presence of edema. Linear Discriminant Analysis (LDA) was employed for feature reduction, and a backpropagation neural network optimized by a genetic algorithm (GA-BP) was constructed for disease prediction. Test results demonstrated high predictive accuracy, reaching 95.3% for stroke, 69.9% for heart disease, and 68.5% for cirrhosis, indicating robust model performance. Furthermore, random forest algorithms were applied to analyze disease comorbidity probabilities, revealing mechanisms of shared risk factors. Sensitivity analysis was conducted to identify key features influencing model outputs. Based on the findings, several optimized prevention strategies are proposed to the World Health Organization, providing a theoretical foundation and methodological support for precise prediction and scientific management of major diseases.