HEART-PREP: Holistic Ensemble-based Algorithm for Robust Transformation and Preprocessing of Heart Disease Datasets

Background: Heart disease is one of the leading causes of morbidity and mortality worldwide, thus early detection and precise prediction are critical for efficient treatment and prevention. Attaining excellent predictive performance is strongly dependent on data quality, which is frequently hampered by missing values, outliers, and irrelevant features. While existing preprocessing approaches attempt to tackle these difficulties, they frequently have shortcomings like as bias, slowness, or less generalizability across datasets. Objectives: This study seeks to (1) create an effective preprocessing framework that incorporates dynamic feature selection, imputation, outlier removal, and scaling methods to enhance dataset quality; (2) improve the performance of machine learning classifiers such as SVM, KNN, DT, and RF in predicting heart disease; and (3) compare the suggested method to existing methods to show its dependability, scalability, and contribution to early heart disease detection. Methods: This study presents HEART-PREP, an ensemble-based preprocessing framework for heart disease prediction, which has been tested on five datasets (Cleveland, Hungary, Long Beach, Switzerland, and Statlog). Clinical criteria include age, gender, chest discomfort, blood pressure, cholesterol, ECG findings, heart rate, and thalassemia were evaluated. The framework includes RACE for outlier elimination, EMERALD for missing value imputation, Min-Max scaling for normalisation, and DYNAMIC for feature selection. Classification was done with SVM, KNN, DT, and RF, and the results were compared to existing techniques in terms of accuracy, precision, recall, and F1-score. Finding: HEART-PREP outperformed SVM (92.26%), KNN (91.07%), DT (92.86%), and RF (95.24%) in accuracy, recall, and F1-score. Novelty: HEART-PREP is unique in that it combines dynamic feature selection, powerful preprocessing, and scaling strategies to consistently exceed prior approaches with up to 95.24% accuracy in heart disease prediction. Keywords: Preprocessing, Imputation, Outlier removal, Feature selection, Ensemble-based algorithms