Optimasi Kinerja XGBoost Dan Random Forest Dalam Klasifikasi Penyakit Hati Berdasarkan Data Klinis Pasien Pada Imbalance Dataset

Abstract

Liver disease is a major health concern that requires early detection through accurate clinical data analysis. Machine learning can support liver disease classification; however, an imbalanced dataset may cause a model to favor the majority class. This study aims to optimize and compare the performance of Random Forest and Extreme Gradient Boosting (XGBoost) algorithms for liver disease classification using clinical patient data. The dataset is the Liver Patient Dataset, comprising 27,158 records and 11 attributes, including the target variable Result. The research methodology includes data preprocessing, an 80:20 training-test split, the application of the Synthetic Minority Over-sampling Technique (SMOTE) to the training data, and hyperparameter optimization. Six experimental scenarios are designed to compare both algorithms before and after SMOTE and parameter optimization. Model performance is evaluated using accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrices. Based on the reported simulation results, XGBoost combined with SMOTE and hyperparameter tuning achieves an accuracy of 89.12%, precision of 88.45%, recall of 88.21%, an F1-score of 88.33%, and an ROC-AUC of 91.27%. These results indicate the potential of combining data balancing and parameter optimization to improve classification performance. Nevertheless, these figures require verification through actual computational experiments before they can be considered empirical findings. This study is expected to provide a basis for evaluating clinical-data-based liver disease classification models and for further research into reliable machine learning approaches for healthcare applications.