MACHINE LEARNING-BASED PREDICTION OF HEART FAILURE USING GENETIC DATA

Abstract

Heart failure is a major global concern affecting millions of people. The disease is characterized by high mortality and significant economic burden. Therefore, in this study, we propose a highly accurate, rapid and timely model for the diagnosis of preclinical heart failure based on genetic biomarkers. This model consists of a Random Forest classifier and 10 differentially expressed genes selected using the particle swarm optimization algorithm. Our results demonstrated its effectiveness, with accuracy, precision, specificity, recall, F1-score, and AUC achieving 99.58%, 99.52%, 97.14%, 100%, 99.76%, and 98.57%, respectively, on the training set, which includes two gene dataset, namely GSE5406 and GSE3586. The test results on the GSE57345 dataset were 97.53%, 98.09%, 89.17%, 99.05%, 98.56%, and 94.11%, respectively. These findings indicate that, despite differences among patient groups, our model remains highly effective and can be applied for personalized disease prediction and precision medicine.