Understanding and predicting crime patterns is essential for enhancing public safety and supporting data-driven decision-making in law enforcement. This study investigates the use of machine learning algorithms to classify types of criminal offenses based on structured data derived from municipal police blotter records. The features used in the analysis include incident type, type of place, barangay, day of the week, month, and detailed demographic attributes of both victims and suspects such as sex, age group, occupation, and status. After preprocessing and feature selection using Principal Component Analysis (PCA), ten offense types with sufficient representation were retained. Five machine learning models—LightGBM, CatBoost, XGBoost, Random Forest, and MLPClassifier—were trained and assessed based on metrics such as accuracy, precision, recall, F1-score, and Cohen’s Kappa. Among them, LightGBM achieved the best performance with an accuracy of 0.82 and a Kappa score of 0.76, closely followed by CatBoost and XGBoost. The results highlight the predictive value of incident-related and demographic features in crime classification tasks. Overall, the findings confirm the strong performance of ensemble-based machine learning models in uncovering crime trends and provides actionable insights that can inform proactive policing strategies.