Geospatial Artificial Intelligence for Safer Cities: Mapping and Predicting the Geography of Crime in India

Abstract

Predictive policing and geospatial artificial intelligence (GeoAI) are being increasingly used to analyze crime geography, yet evidence from India remains fragmented across geography, criminology, computer science, and public administration. This systematic review addresses fragmentation by synthesizing 29 evidence units drawn from 46 publications, including one sustained 18-publication single-city program, and by situating Indian evidence within international GeoAI and spatial-criminology research. Existing approaches are limited by the predominance of retrospective, single-city hotspot mapping; heterogeneous and noncomparable evaluation metrics; limited independent validation of predictive systems; weak reporting of explainability and data provenance; and minimal integration of environmental and climate variables. A PRISMA-informed search of six databases, supplemented by citation and snowball tracking, was used to identify, screen, and appraise the evidence. The synthesis compares kernel density estimation, Getis-Ord Gi*, Moran’s I, risk-terrain modeling, and machine-learning/deep-learning approaches across local, district, state, national, and international contexts. The review reveals that the spatial concentration of crime is consistently reported in Indian studies that test for clustering, whereas a recent Indian machine-learning study reports 93.20% accuracy but lacks independent replication and sufficiently harmonized validation. The Delhi Crime Mapping, Analytics and Predictive System (CMAPS) case also demonstrates how reporting and data-provenance mechanisms can shape place-based predictive risk. The principal contribution is an integrated evidence framework that separates descriptive spatial evidence from predictive claims, makes the study-selection and quality-assessment procedures explicit, and identifies six research priorities: harmonized data infrastructure, bias auditing, place-based modeling, climate-crime integration, cross-state benchmarking, and independent evaluation. These findings provide a methodological basis for developing more transparent, reproducible, and context-sensitive GeoAI research on crime in India.