Differential Evolution Wrapper-Based Feature Selection Method for Stroke Prediction

Stroke is a medical condition in which a blood vessel in the brain bursts and causes brain damage. After realizing the wide-ranging effects, a brain infarction can have on a community, significant efforts have been made to enhance stroke therapy and diagnosis. The medical practitioner can benefit from more precise diagnosis if the occurrence of strokes can be predicted using patient medical records. Identifying the important features decreases the number of features by eliminating irrelevant or misleading, noisy and redundant data which can accelerate the process of prediction. The proposed work, involves the wrapper-based differential evolution approach for best feature selection and based on best features performing the stroke prediction using random forest algorithm. To evaluate the performance of proposed work, wrapper-based sequential forward, backward, and wrapper-based built around random forest algorithm are evaluated. Using a wrapper-based feature selection method, this work identifies critical characteristics, and then proposes stroke prediction accuracy estimation. The following are the findings from our investigation into methods for selecting features that make use of proposed wrapper-based differential evolution algorithm. The most significant features are age, ever married, residence type, and Avg_glucose_level. The proposed prediction model achieved maximum accuracy of 95.79%.