Strategi Optimalisasi Hyperparameter Model Machine Learning untuk Prediksi Putus Studi Dini Mahasiswa
Abstract
One of the biggest problems facing institutions of higher learning is the phenomenon of student dropout. The purpose of this study is to optimize machine learning models in order to enhance the performance of student dropout prediction models. Three hyperparameter tuning (HT) techniques—Grid Search Cross-Validation, Randomized Search, and Bayesian Optimization—were used to assess six machine learning (ML) classification models: K-Nearest Neighbors, Decision Tree, Random Forest, Gradient Boosting, AdaBoost, and Support Vector Machine. Macro F1-score, macro recall, and macro precision—all pertinent for unbalanced datasets—were used to gauge the model's performance. According to experimental findings, HT continuously enhances model performance above the baseline. While Randomized Search also works well for some models, Bayesian Optimization typically produces competitive solutions with good efficiency. This study sheds light on the best HT techniques for predicting student dropout.