Corresponding Author's : Sneha

Abstract :

The requirement for crime prevention and prediction continues to grow with regard to the manner in which urban governments are governed. Law enforcement agencies have evolved from being reactive in response to criminal incidents to proactive by attempting to anticipate or prevent those events. This paper outlines a large-scale computational approach using over 40,000 criminal record data points from 29 cities across India (2020-2024) to support an "anticipate and prevent" paradigm. A dual-pathway analytical framework is utilized to address the complexity of working with real-world data. The first pathway utilizes a SARIMAX model with a one-year "cold-start," to abstract stable long-term trends within the raw data. The second pathway incorporates a high-precision hybrid ensemble utilizing Random Forest and Gradient Boosting to assess multi-dimensional characteristics (city encoding/demographics), which demonstrated a mean absolute error of 0.24. Additionally, an optimized version of Agglomerative Clustering with Silhouette Denoising was utilized to categorize spatial risk profiles of the cities and to identify new hotspot locations. The utilization of this framework provides a bridge from the theoretical application of machine learning to its practical application. Therefore, it provides a validated methodology to perform both robust noise-resilient regional analyses and accurate high-fidelity local predictions. Overall, the results presented herein provide authorities with the capability to utilize these data to inform resource allocation strategies that will enable their continued effectiveness in preventing crime as the evolving nature of criminal behavior dictates.

Keywords :

Crime Prediction, SARIMAX, Hybrid Ensemble, Clustering, Machine Learning, Hotspot Identification

Article Details

Unique Paper ID: 61001

Publication Volume & Issue: VOLUME 6 - 2026, ISSUE 1

Page(s): 1-15

Article Preview & Download

Share This Article

• X