Heart Disease Prediction Using a Hybrid Feature Selection and Ensemble Learning Approach

The main objective of this research is to design and develop a robust predictive framework for heart disease classification by integrating hybrid feature selection and ensemble learning techniques. The study aims to preprocess the dataset effectively through normalization, encoding, and handling of missing values to ensure data consistency and model reliability. A hybrid feature selection process combining Random Forest and Recursive Feature Elimination (RFE) is implemented to identify the most influential clinical parameters while eliminating redundant and noisy features. The refined dataset is used to train multiple classifiers such as CatBoost, LightGBM, and Deep Neural Networks (DNN) to evaluate individual model performance. To further enhance prediction accuracy and stability, an ensemble model integrating CatBoost, DNN, and AdaBoost is developed, where the ensemble weights are optimized using a Genetic Algorithm. The objective is to achieve high accuracy, precision, recall, and F1-score while maintaining interpretability and computational efficiency. This research ultimately seeks to establish an adaptive and scalable classification model capable of improving diagnostic reliability and decision-making within structured medical datasets.