Optimizing heart disease prediction through ensemble and hybrid machine learning techniques

Abstract

In this modern era, heart diseases have surfaced as the leading factor of fatalities, accounting for around 17.9 million lives annually. Global deaths from heart diseases have surged by 60% over the last 30 years, primarily because of limited human and logistical resources. Early detection is crucial for effective management through counseling and medication. Earlier studies have identified key elements for heart disease diagnosis, including genetic predispositions and lifestyle factors such as age, gender, smoking habits, stress, diastolic blood pressure, troponin levels, and electrocardiogram (ECG). This project aims to develop a model that can identify the best machine learning (ML) algorithm for predicting heart diseases with high accuracy, precision, and the least misclassification. Various ML techniques were evaluated using selected features from the heart disease dataset. Among these techniques, a combination of random forest (RF), multi-layer perceptron (MLP), XGBoost, and LightGBM employing an ensemble method with a stacking classifier, along with logistic regression (LR) as a metamodel, achieved the highest accuracy rate of 95.8%. This surpasses the efficiency of other techniques. The suggested method provides an encouraging framework for early prediction, with the overarching goal of reducing global mortality rates associated with these conditions.