Improving the Efficiency of Predicting the Heart Diseases Using Optimized Feature Selection and Ensemble Machine Learning Techniques

Abstract

Millions of people worldwide suffer from heart failure, a chronic illness that makes an effective machine learning (ML)-based approach for early detection and treatment necessary. Although medication is still the mainstay of care, exercise is becoming recognized as a useful adjunctive therapy for the management of heart failure. In this work, we used patient health parameter data to design a ML-based method to enhance heart failure detection. Improving the early detection of heart failure is our goal in an effort to save lives. To find the most important features for enhancing performance, we conducted a comparative analysis of ten distinct ML algorithms and applied feature engineering methodologies. By developing a novel new feature set, we improved our strategy and obtained the best accuracy ratings. The proposed system works on the statistical dataset and CT scan images. Numerous experiments were carried out to assess the efficacy of different algorithms, and our suggested approach outperformed other cutting-edge models, attaining impressive accuracy. Cross-validation approaches were employed to validate all applied procedures. On the CT scan dataset, AdaBoost (AB) achieved 100% accuracy, while gradient boosting (GB) led with 96% on the statistical dataset. Accuracy improved with random or synthetic data. Notably, applying a soft voting ensemble of all models further boosted accuracies to 98% and 95% on the respective datasets. Our study advances heart failure early detection techniques, which make important scientific contributions to the medical world.