Robust Heart Disease Prediction: A Hybrid Approach to Feature Selection and Model Building

This research work integrates advanced data preparation methods and advanced modelling approaches to offer a complete approach for creating a predictive model for heart disease. Data cleansing starts the procedure in order to fix disparities and manage missing values, thereby guaranteeing the dependability and accuracy of the dataset. By producing synthetic samples for the minority class, the Synthetic Minority Over-sampling Technique (SMOTE) solves class imbalances and improves the accuracy of the model to forecast minority class occurrences. The z-score approach helps to control outliers thereby preventing distortion in the study. Using a hybrid Filter-Recursive Feature Elimination (FRFE) approach-which combines filter and wrapper techniques to find the most pertinent features, therefore lowering multicollinearity and boosting model performance-feature selection is accomplished. Important characteristics found are age, kind of chest discomfort, maximal heart rate attained, ST depression brought on by exercise, and fluoroscopy-induced coloration of several major arteries. Using AdaBoost, an ensemble learning technique combining several weak classifiers to create a strong one, the improved dataset is utilised to develop a predictive model. The resultant model shows great accuracy (93%), precision (93%), recall (94%), and an amazing ROC AUC score (0.98), therefore highlighting its potential use in healthcare environments and heart disease prediction ability.