<![CDATA[Online games have evolved into digital social spaces where player interactions often include toxic communication, potentially affecting user experience and psychological well-being, especially among younger players. This research is intended to examine and compare the performance of various machine learning algorithms in classifying toxic chat on the Roblox platform and to identify underlying linguistic patterns. The dataset consists of 7,119 Indonesian-language chat data labeled into six categories: identity_hate, insult, obscene, severe_toxic, threat, and toxic. The methodology includes data preprocessing, text representation using Bag-of-Words (BoW) and TF-IDF, and classification using Naive Bayes, Support Vector Machine (SVM), and Random Forest. To assess how well the model performs, several metrics are used, including accuracy, precision, recall, F1-score, and 3-fold cross-validation. The results show that SVM with TF-IDF achieves the best performance with 84.48% accuracy, followed closely by SVM with BoW. The findings indicate that while classical machine learning models remain effective, challenges persist in distinguishing linguistically similar categories.]]>
Copyrights © 2026
