Implementasi Model XGBoost dengan Hyperparameter Tuning Grid Search pada Sistem Klasifikasi Gangguan Tidur Berbasis Web

Abstract

Sleep disorders can impact physical health, mental health, and individual productivity. The high prevalence of sleep disorders indicates the need for early detection efforts to help reduce the risk of developing various chronic diseases. However, conventional diagnostic methods such as Polysomnography (PSG) still have limitations due to high costs and long examination times. The application of machine learning technology can be utilized to support the early detection process of sleep disorders more quickly and efficiently. This study aims to develop a sleep disorder classification system using the Extreme Gradient Boosting (XGBoost) algorithm with the implementation of Grid Search hyperparameter tuning and the SMOTE oversampling technique to address data imbalance. This study used secondary data from the Sleep Health and Lifestyle Dataset, consisting of 374 records with 13 features, including age, gender, sleep duration, sleep quality, physical activity level, stress level, BMI category, blood pressure, heart rate, and daily steps. The research stages included data preprocessing, feature engineering, categorical data encoding, data splitting using 70:30 and 80:20 scenarios, SMOTE implementation, and the classification process using the XGBoost algorithm. The results showed that the best scenario was obtained using the 80:20 data split with hyperparameter tuning, resulting in the best parameters: colsample_bytree = 0.7, learning_rate = 0.1, max_depth = 5, min_child_weight = 1, n_estimators = 100, and subsample = 0.8. The model achieved an accuracy, precision, recall, and F1 score of 0.97. These results indicate that the combination of XGBoost, Grid Search, and SMOTE can improve the performance of sleep disorder classification effectively.