Salahudin Robo, Triyanna Widiyaningtyas, Wahyu Sakti Gunawan Irianto, Wiwik Suharso, Muhammad Fatkhur Rizal, Rahmawati Febrifyaning Tias, Mas Nurul Hamidah, Erfan Ainul Yakin
Collaborative filtering is one of the most used techniques in recommendation systems. This technique generates items by analyzing similarities between users based on their preferences and behavior. When recommending items to users, there is one final step that is crucial, namely, predicting the ratings that users have not yet given to specific items. A technique often used in final rating prediction is Gradient Boosting. This study aims to compare and evaluate the performance of three gradient boosting algorithms, namely LightGBM, XGBoost, and CatBoost. The datasets used are MovieLens 1M and 100K, which have high sparsity levels. Model performance is evaluated using two primary metrics: Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The results show that the XGBoost algorithm performs best on all datasets. On MovieLens 100K, XGBoost obtained an RMSE value of 0.9413 and an MAE value of 0.7446, while LightGBM obtained an RMSE value of 0.9427 and an MAE value of 0.7462, and CatBoost obtained an RMSE value of 0.9487. MAE 0.7515. Meanwhile, on the MovieLens 1M dataset, XGBoost achieved an RMSE value of 0.9051 and an MAE of 0.7148, while LightGBM obtained an RMSE value of 0.9124 and an MAE of 0.7215. The CatBoost algorithm achieved an RMSE value of 0.9135 and an MAE of 0.7222. This finding confirms that choosing the correct algorithm can have a significant impact on the performance of recommendation systems. © 2025 IEEE.
Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia; Universitas Muhammadiyah Jember, Department of Information Systems, Jember, Indonesia; Universitas Hasyim Asy'ari, Department of Information Technology, Jombang, Indonesia; Universitas Bhayangkara Surabaya, Department of Informatics, Surabaya, Indonesia
