<![CDATA[Crime data management at the Bengkulu City Police Station currently faces several operational constraints, including underutilized historical data and conventional analytical methods that prolong the crime-pattern mapping process. This study aimed to analyze crime patterns through a data science approach using the CatBoost Classifier method to support data-driven tactical decision-making within the police department. A quantitative methodology using an Applied Data Science framework was executed following the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework. The study analyzed crime report data extracted from the Bengkulu City Police database, focusing on temporal (time of occurrence) and spatial (district/kecamatan) variables. Model performance was comprehensively evaluated using a Confusion Matrix. Findings revealed that Motor Vehicle Theft (Curanmor) was the most dominant crime category, with peak vulnerability observed in Selebar District and the majority of incidents concentrated during the early morning hours. Feature Importance analysis indicated that the temporal dimension (Time Category) contributed the most (30.37%), followed by geographical location (District) at 29.30% identifying them as the primary predictors shaping crime patterns. Model evaluation for the CatBoost Classifier yielded an overall accuracy of 49%, reflecting performance limitations due to the small dataset size and imbalanced class distribution. Nevertheless, the model demonstrated a high recall value (89%) specifically in detecting Motor Vehicle Theft. maintaining its mility as an early warning signal for street crimes. Conversely, the model completely failed to predict fraud offenses (0% accuracy) because fraud characteristics are not tied to spatial-temporal patterns. The application of the CatBoost Classifier in this study ts best suited as a tactical intelligence support instrument for street crime mitigation, enabling police administrators to optimize directed patrol strategies in high-risk zones and time windows]]>
Copyrights © 2026
