Analisis Trade-off Pendekatan Greedy dan Metaheuristic dalam Seleksi Fitur Terhadap Model Ensemble

Abstract

The increasing volume and dimensionality of medical data pose challenges for effective machine learning model development. Feature selection techniques (FST) are crucial for improving model performance, computational efficiency, and interpretability. This study analyzes the trade-off between greedy and metaheuristic FST approaches in optimizing Decision Tree-based ensemble models. We compare Mutual Information-Sequential Backward Selection (MI-SBS) as a greedy method and Binary Grey Wolf Optimization (BGWO) as a metaheuristic method. FST fitness is evaluated using a Decision Tree Classifier with 5-fold cross-validation. Final classification performance is assessed using AdaBoost and XGBoost on three distinct medical datasets. Results indicate that MI-SBS offers faster feature selection and stable accuracy, often outperforming the baseline. BGWO, while slower in selection, achieves greater feature reduction, leading to significantly faster final model training at the cost of a minor accuracy decrease. This research provides insights into selecting appropriate FST based on desired trade-offs between computational efficiency and classification accuracy in health informatics.