The rapid growth of the mobile gaming industry in Indonesia, particularly Mobile Legends: Bang-Bang, has generated millions of user reviews on the Google Play Store, making manual analysis inefficient and prone to bias. This study compares three algorithms—Naive Bayes, Support Vector Machine (SVM), and Logistic Regression—for sentiment analysis of 52,651 reviews. Preprocessing includes text cleaning, stopword removal (Indonesian/English), Sastrawi stemming, and TF-IDF representation (min_df=3, max_df=0.9, n-gram 1–2). Binary labeling follows a rating-based approach: 1–2 stars (negative), 4–5 stars (positive), while 3-star reviews are excluded due to ambiguity. Evaluation using accuracy, precision, recall, F1-score, confusion matrix, and Cohen’s Kappa shows SVM and Logistic Regression achieving ≈90–91%, with SVM chosen as the default model for its balanced metrics and margin stability. The model can be deployed as an API service (Flask/FastAPI) for near real-time review monitoring (e.g., lag, AFK, matchmaking), enabling alert thresholds and improvement prioritization. Findings remain limited to Mobile Legends reviews on Google Play, requiring further validation across other applications.