Hybrid Feature Selection and Ensemble Learning for Multi-Disease Prediction in Smart Healthcare Systems

Abstract

Smart healthcare systems with high dimensional clinical data, feature redundancy and limited generalization of conventional models pose a challenge to the accurate prediction of multiple diseases. The main goal of this study is to overcome these problems by introducing a hybrid feature selection and ensemble learning framework which improves the prediction accuracy and also has better computational efficiency. The method is based on two-stage feature optimization process, which combines ReliefF to rank the features as relevant and a Genetic Algorithm to determine the optimal subset that will have the best effect as a reduced dimensionality of the feature space while retaining essential information. The optimized features are used in a weighted ensemble model (RF, SVM and XGB) to boost the robustness of the classification. The framework is tested on the benchmark UCI Heart Disease Dataset for a standardized test. Experimental results show that the accuracy is 97.48%. The precision, recall and F1-score are 96.72%, 97.05% and 96.88% respectively which are better than the baseline models and standalone models. The main idea is to adaptively fuse the hybrid feature optimization and ensemble intelligence, which effectively avoids overfitting and improves the generalization of the model. The proposed approach is scalable and can be applied in real-time clinical decision making which is a reliable approach towards intelligent disease prediction in a smart healthcare environment