PREDIKSI RATING FILM DENGAN PERBANDINGAN ALGORITMA CATBOOST DAN RANDOM FOREST

Abstract

The increasingly rapid development of the film industry has caused the number of films produced and published to continue to rise, resulting in an ever-growing volume of available film data. This condition drives the need for methods capable of utilizing film metadata to predict rating values. This study aims to build a film rating prediction model and compare the performance of the CatBoost and Random Forest algorithms using the TMDB 5000 Movies dataset obtained from Kaggle. The initial dataset consists of 4,803 film records with vote_average as the target variable and eight predictor attributes, namely budget, runtime, release_year, genres, original_language, main_country, main_language, and production_company. The research stages include data quality checking, preprocessing, JSON attribute extraction, feature engineering, categorical data transformation, an 80:20 data split, modeling, hyperparameter tuning, evaluation using MAE, RMSE, and R², as well as 5-fold cross-validation. Random Forest uses One-Hot Encoding for the retained categorical features, while CatBoost processes certain categorical features natively. The hold-out results show that CatBoost achieved an MAE of 0.6596, an RMSE of 0.9727, and an R² of 0.3725, whereas Random Forest achieved an MAE of 0.6604, an RMSE of 0.9778, and an R² of 0.3660. In the 5-fold cross-validation, Random Forest achieved a Mean MAE of 0.6709, a Mean RMSE of 0.9942, and a Mean R² of 0.3061, while CatBoost achieved a Mean MAE of 0.6719, a Mean RMSE of 0.9948, and a Mean R² of 0.3048. Thus, the two algorithms have relatively comparable performance; CatBoost is slightly better on the hold-out, while Random Forest is slightly better on the cross-validation. Both models were then integrated into a web application using Flask as the machine learning service and CodeIgniter 4 as the interface.