Comparative Analysis of Supervised Learning Algorithms in Indonesian Film Genre Classification Based on the CRISP-DM Approach

Abstract

This research is motivated by the complexity of the classification of Indonesian film genres due to language variations, inconsistent synopsis structures, and similarities in characteristics between genres. Given the limited comparative studies on local data, this study proposes an analysis of various supervised learning algorithms. The goal is to evaluate and determine the most optimal model for accurately classifying film genres based on synopsis texts.

This contribution uses the CRISP-DM framework to compare Decision Tree, Naive Bayes, SVM, and KNN algorithms. Data in the form of synopsis of Indonesian films is processed through text cleaning and word-frequency-based feature extraction. His main contribution is to provide a systematic evaluation and a replicable framework for the development of Natural Language Processing (NLP) in the Indonesian film industry.

The results showed that SVM provided the highest accuracy, while Naive Bayes excelled in time efficiency. Although Decision Tree is easy to interpret and KNN is sensitive to parameters, the performance of both can be significantly improved through normalization techniques as well as feature selection. The confusion matrix evaluation noted misclassification in drama and comedy genres, but the results of cross-validation ensured that the model remained consistent and stable across various data divisions.

Overall, this study shows that the CRISPDM approach can be used well in the process of classifying Indonesian film genres. Based on the results obtained, the Support Vector Machine is recommended as the best algorithm for this case because it is able to provide the most optimal performance compared to other methods.