<![CDATA[Selecting the right hyperparameter tuning method plays a crucial role in improving the performance of a classification model, especially when applied to datasets with different class distribution characteristics. This study aims to analyze and compare the effectiveness of three hyperparameter tuning methods, namely Grid Search, Random Search, and Bayesian Optimization, on the XGBoost, Random Forest, and Support Vector Machine (SVM) models. Testing was conducted using two datasets with different characteristics, namely Breast Cancer Wisconsin as a balanced dataset and Credit Card Fraud Detection as an unbalanced dataset. Model performance evaluation was adjusted to the characteristics of the datasets, using F1-score (macro) for the Breast Cancer dataset and Precision-Recall AUC for the Credit Card Fraud dataset. The results show that on balanced datasets, all tuning methods produce relatively similar performance, with SVM consistently providing the best results. Conversely, on unbalanced datasets, random search and Bayesian optimization methods show superiority in finding hyperparameter configurations that can improve the detection ability of minority classes, especially on the XGBoost model. This finding emphasizes that the selection of tuning methods and evaluation metrics must be adjusted to the characteristics of the data used.]]>
Copyrights © 2026
