Enhancing XGBoost and CatBoost Methods for Diagnosing Parkinson's Disease Through the Integration of SMOTE and Feature Selection Techniques

Parkinson's disease is a neurodegenerative condition affecting movement, requires early detection for effective treatment. The main symptoms of this disease include tremors, muscle stiffness, slow movements, and difficulty controlling movements. Parkinson's disease that is successfully detected early can be given effective medical treatment. Recent studies suggest that machine learning can be a useful indicator for diagnosis. However, predicting Parkinson's disease remains challenging due to imbalanced data. Utilizing feature selection, imbalance correction techniques, and optimization can enhance prediction accuracy. The results of this study show that XGBoost consistently performs better than CatBoost in initial scenarios across various data split ratios. Both XGBoost and CatBoost achieved their highest accuracy at an 80:20 data split ratio, with XGBoost reaching 91.39% and CatBoost 90.72%. Without hyperparameter tuning through GridSearchCV, CatBoost achieved its highest accuracy of98.01 % using SMOTE and FS3, while XGBoost achieved its highest accuracy of 96.68% with SMOTE and FS1. The application of GridSearchCV significantly improved both accuracy and F1 Score for CatBoost and XGBoost. With GridSearchCV, both models demonstrated consistent performance enhancements across all tested scenarios. Overall, these findings highlight the effectiveness of hyperparameter tuning and feature selection in optimizing model performance for complex classification tasks.