Multi-objective nature inspired hybrid optimization algorithm to improve prediction accuracy on imbalance medical datasets

Imbalanced medical datasets pose a significant challenge for predictive modelling. The current study presents a new method of performing feature selection specifically for the imbalanced medical datasets to improve accuracy of the predictions. The proposed Multi-Objective Feature Selection with Cost-Sensitive (MOFSCS)algorithm leverages the large-scale exploration capability of the Squirrel Search to generate diverse candidate feature subsets and employs Tabu Search for local optima refinement. One of the key developments is learning with consideration of costs, which is closer to the identification of the minority class. The effectiveness of the proposed approach is ensured by the experiments on different imbalanced medical datasets, namely, heart disease and stroke prediction datasets. The results reveal that the proposed method, when integrated with the XGBoost classifier, achieves a precision of 98.5%, recall of 98.7%, F1-score of 98.6%, accuracy of 98.7%, and an AUC-ROC of 98.7% on the heart disease dataset. Similarly, for brain stroke prediction, the model attains a precision of 98.9%, recall of 99.0%, F1-score of 98.9%, accuracy of 99.0%, and an AUC-ROC of 99.0%.