Analysis of Clustering Results for Crime Incident Data in Indonesia Using Fuzzy C-Means

Abstract

This study examines the clustering of crime incident data across Indonesia from 2000 to 2024 using the Fuzzy C-Means (FCM) algorithm, with a focus on the impact of data normalization. Comprehensive annual provincial crime statistics from Badan Pusat Statistik (BPS) were preprocessed to handle missing values and then standardized via the Standard Scaler. FCM clustering was performed separately on both the original and normalized datasets, with the number of clusters set to three. Cluster quality was evaluated over ten independent runs using five metrics: Davies-Bouldin Index (DBI), Silhouette Score (SS), Calinski-Harabasz Index (CH), Adjusted Rand Index (ARI), and Normalized Mutual Information (NMI). Results indicate that normalization consistently yields lower DBI values (average 0.824 vs. 0.830) and higher SS (average 0.367 vs. 0.363) and CH scores (average 55.35 vs. 54.09), while ARI and NMI remain stable across treatments. These findings demonstrate that normalization enhances cluster compactness and separation without altering underlying data structures, leading to more interpretable and reliable groupings. By uncovering regional crime patterns and highlighting the methodological importance of preprocessing, this research provides actionable insights for policymakers and law enforcement agencies to allocate resources more effectively and develop targeted crime prevention strategies.