Analysis of Text Mining Clustering on Suara Surabaya Crime Report with DBSCAN Neural Network Autoencoder Algorithm

Abstract

Criminality, or crime, is a behavior that violates the law or is contrary to applicable values and norms. A high number of criminal behaviors criminal behaviors in a community significantly impacts its social conditions, leading to a decrease in welfare, unrest, and material losses that pose a threat to an individual's life. This study examines text mining on crime report data from Suara Surabaya using the DBSCAN clustering method and the Neural Network Autoencoder. The neural network autoencoder algorithm effectively reduces the data dimension, with an input dimension of 300 and an encode dimension of 64. Clustering analysis using the DBSCAN method based on the silhouette coefficient value criterion resulted in three clusters, with cluster 1 dominating the report. The clustering results show essential patterns in complaint reports, and LDA analysis reveals critical topics in the report. Cluster 0 shows a diversity of reports focusing on motor loss, interaction with homes or properties, and people's entry into homes. Cluster 1 is more focused on the loss of vehicles, both cars and motorcycles, with specific details such as vehicle color, number, brand, and related transactions or social interactions. Meanwhile, cluster 2 focuses on reports related to interactions with police stations and information on the location of incidents. This text mining approach to community crime report data not only improves analysis accuracy and efficiency, but also provides essential information that can support efforts to handle and prevent crime.