Cardiovascular disease prediction: Ranking the significant features using Hybrid Accumulated Feature Selection (HAFS) Method

Abstract

Heart disease has become a major problem recently, lowering people’s standard of living. There is a pressing need to improve prediction models for cardiac data, machine learning has achieved outstanding results in predicting and decision-making. To test the suggested model, this research makes use of the heart disease dataset, which has more than 70,000 records. Body Mass Index, Mean Arterial Pressure, and Pulse Pressure are three additional features that have been enhanced to the dataset in order to enhance the performance. For the most important feature selection, this research suggests the HAFS (Hybrid Accumulated Feature Selection) model. The HAFS design incorporates three statistical methods: Mutual Information (MI), the ANOVA f-test, and the Chi-squared test. The investigation is conducted with the use of various ML and DL classification algorithms, including SVM, NB, LR, XGBoost, LGBoost, AdaBoost, Stochastic gradient descent, and ANN. The experimental findings show that ANN and XGBoost are the best.This work highlights the crucial importance of feature engineering and hyperparameter adjustment in enhancing the accuracy of predictive models.These findings support the ongoing endeavours to create dependable and efficient instruments for the early identification and intervention of cardiac disease.Investigating advanced feature selection techniques and hyperparameter optimization methods can further enhance model performance.