Empirical analysis of predicting heart disease using diverse datasets and classification procedures of machine learning

Abstract

Cardiovascular disease (CVD) poses a significant threat due to its complexity and fatality, necessitating early intervention. Fortunately, the rapidly evolving field of machine learning (ML) offers an array of algorithms for disease diagnosis and prediction. This research aims to develop and identify a model that assists radiologists in predicting heart disease, a two-phased approach. Phase 1: Feature Selection with SelectKBest: The first phase utilizes the SelectKBest method to select the most relevant features for prediction. This method combines individual feature rankings based on three statistical tests: chi-squared, mutual information, and F-statistic. The final selection is based on the overall rank obtained by each feature. Phase 2: Classification Algorithm Exploration: The second phase applies various classification algorithms, including Random Forest, k-nearest neighbors (KNN), decision tree (DT), support vector machine (SVM), Naïve Bayes (NB), Logistic Regression (LR), Random Forest Grid search, Gradient Boost, and Neural Network. The performance of these models is evaluated using three standard heart disease datasets: Cleveland, Faisalabad, and Framingham, retrieved from UCI and Kaggle. Each dataset undergoes pre-processing before applying the SelectKBest feature selection and all ML algorithms. Across all three datasets, Random Forest emerged as the champion, achieving accuracy rates of 90.16%, 90%, and 84%, respectively. Additionally, it demonstrated consistently lower classification errors compared to other algorithms. This research highlights the effectiveness of feature selection, particularly the SelectKBest filter-based method, in improving heart disease prediction accuracy using machine learning models like Random Forest. This paves the way for integrating such models into clinical settings, empowering radiologists with valuable decision-making tools for early CVD detection and intervention.