Perbandingan Logistic Regression dan Random Forest untuk Prediksi Risiko Stroke

Abstract

Stroke is a leading cause of global mortality that requires early detection through artificial intelligence technologies. This study aims to conduct a comparative analysis between Logistic Regression and Random Forest models in predicting stroke risk based on health and lifestyle data. The study utilized the Stroke Prediction Dataset, which underwent preprocessing, standardization, and class imbalance handling. Model performance was evaluated using Accuracy, Precision, Recall, and ROC-AUC metrics. The results demonstrate that Random Forest significantly outperformed Logistic Regression with an AUC value of 0.990 compared to 0.892. Random Forest proved to be more effective for medical screening, achieving a Recall of 96.29%, which is substantially higher than Logistic Regression at 83.71%. Furthermore, Random Forest demonstrated superior capability in capturing non-linear data patterns in complex variables such as BMI. It is concluded that Random Forest is the recommended method for stroke early detection systems due to its ability to minimize undetected positive cases (false negatives).