OPTIMISASI PEMILIHAN FITUR UNTUK PREDIKSI GAGAL JANTUNG: FUSION RANDOM FOREST DAN PARTICLE SWARM OPTIMIZATION

Abstract

Heart failure is a serious, life-threatening cardiovascular disease that increases with age and unhealthy lifestyles. Early prediction is essential to provide timely treatment and reduce mortality. The use of machine learning techniques, especially the Random forest (RF) method, for predicting heart failure has been previously researched, so the problem that occurs is that the RF method does not have maximum results because of irrelevant features. Selection of relevant features is a key step in building an accurate prediction model. Particle Swarm Optimization (PSO) is used to improve feature selection by searching for optimal combinations. The aim of the research is to reduce the mortality rate by improving the RF method with relevant features so as to increase the accuracy of predictions with Fusion RF and PSO. The results show an increase in accuracy of 02.78% to 87.33% with PSO, although the AUC decreased by 0.031%. The advantage of PSO is a significant increase in accuracy, but the disadvantage is a slight decrease in AUC. Future developments could explore how to address AUC degradation without compromising accuracy and transmitting additional relevant features.