Predicting Stroke Risk Using Random Forest Classifier: A Healthcare Data Analysis

This work explores the Random Forest classifier's effectiveness in analyzing healthcare data for predicting stroke risks. In this study, data preprocessing is done intensively, which ranges from handling missing values, encoding categorical variables, and handling class imbalance, where the data is oversampled using SMOTE to balance the data between stroke and no-stroke cases. After preprocessing, the proposed approach uses the Random Forest classifier to predict the possibility of a stroke in people for the processed data. Further, features related to model selection and characteristics, including a grid search that incorporates cross-validation, are used to fine-tune the classifier for the best results. The performance of the developed Random Forest model is evaluated using accuracy, precision, recall, and F1 score. Moreover, the interpretability of the model is examined to determine which factors affect stroke prediction. The results prove that the Random Forest classifier has high potential in the prediction of stroke risks with 94.42% accuracy, using advanced model selection and tuning techniques. Therefore, this study contributes towards clinical applications employing the Random Forest method, especially when applied with a preprocessing approach such as SMOTE, which can be beneficial in clinical settings for its ability to assess stroke risk mutually with approaches pertinent for healthcare practitioners and academics.