Mapping Crime Dynamics: Integrating Textual, Spatial, and Temporal Perspectives

In recent years, predicting crime for a particular region using Machine Learning (ML) has gained much attention. We applied algorithms to categorize crime types and examine the local vulnerability with hotspot clustering for each crime type for the city of Los Angeles. The combined analysis of classifications with bagging and boosting algorithms and subsequently identifying spatial clusters enhances our comprehension of crime patterns. We further dealt with text-based analysis, employing the Latent Dirichlet Allocation (LDA) model to uncover the latent topics within crime descriptions, followed by clustering these topics using algorithms such as BIRCH, Mini Batch KMeans, OPTICS, and HDBSCAN. As for the temporal aspects of the crime data, we applied Long-Short Term Memory (LSTM) model to analyze fluctuations in the number of crimes and forecast future possibilities. Through this comprehensive study covering various aspects such as spatial, temporal, and textual-based crime dynamics, we can analyze the nature of crimes most likely to occur at specific times and in particular areas. Moreover, through our comparison of various models and algorithms applied, we contribute to the academic discourse in crime analysis. It provides valuable insights into addressing crime reports, spatial clustering, topic modeling, and LSTM applications. It gives practical implications for crime prevention strategies and optimizing resource allocation.