Implementation of Machine Learning Algorithm for Stroke Risk Classification by Applying Sequential Forward Selection

Stroke is a critical medical condition that can abruptly impair brain function. It may arise from either hemorrhaging or obstruction within the blood vessels. Stroke ranks as the second most common cause of death worldwide and is the third primary contributor to disability. This study employs machine learning methodologies, including Random Forest (RF), XGBoost, and LightGBM, to assess the classification of stroke risk. The classification is conducted in two phases: first without feature selection and then utilizing Sequential Forward Selection (SFS) for feature selection. Findings indicate that implementing SFS with an 80:20 data-sharing ratio enhances classification accuracy compared to the approach without feature selection. SFS identifies nine key features: gender, age, hypertension, smoking status, BMI, average glucose level, marital status, employment type, and residence type. Utilizing these nine features achieved the highest accuracy of 95.94%, with a recall of 95.77%, precision of 96.07%, F1-Score of 95.92%, and AUC of 99.32%, outperforming both the Random Forest and LightGBM algorithms. These results demonstrate that the model effectively classifies stroke risk data by focusing on pertinent features.