INVESTIGATION AND COMPARISON OF FEATURE SELECTION TECHNIQUES WITH HYPER PARAMETER TUNING FOR PREDICTING PROACTIVE CARDIOVASCULAR DISEASE

Despite efforts to treat cardiovascular disease (CVD), yet it remains, still, one of the major cause of death; hence, there has been a need for developing accurate and swift predictive models to enable early diagnosis and intervention for such patients. This study analyzes and considers several feature selection models in the presence of hyperparameter tuning to improve CVD prediction models. Feature selection is important in improving model’s interpretability, reduce computational complexity and remove redundant or irrelevant feature. Additionally, which of the filter, wrapper, and embedded methods (e.g. Mutual Information, Recursive Feature Elimination, and LASSO regression) are best for working with CVD dataset are considered. In order to further improve the model performance, Hyper Parameters tuning on machine learning classifier like Logistic Regression, Support Vector Machine, Random Forest and XGBoost using Grid Search and Bayesian optimization are applied. Different feature selection and hyper parameter tuning combination are assessed based on its performance metric including accuracy, recall, precision, F1-score and area under receiver operating characteristic (ROC-AUC) curve. The proposed method has attained 89% which is far better than the models like Logistic Regression (LR) of 85%, Support Vector Machine (SVM) of 81%, K-Nearest Neighbors (KNN) of 86.9%, Artificial neural networks (ANN) of 87%. In particular the proposed method is 4.7% better than LR, 9.8% better than SVM, 2.29% better than KNN and 2.3% better than ANN. The results from experiment show that the performance of model can be improved greatly with the complement of XGBoost and Random Forest when feature selection integrated with optimized hyperparameter tuning. Modern IT methodologies boost efficiency of models by using fewer resources while achieving extended operational life. Feature selection integration enhances diagnostics predictions because IT-based analytics plays an important role in medical diagnosis.