PROBABILISTIC MODELING OF SEQUENTIAL CRIME RISK USING HIDDEN MARKOV MODELS UNDER THE POCSO FRAMEWORK

Abstract

The rapid increase in digital and contact-based offenses against children necessitates predictive, interpretable, and legally defensible analytical frameworks. This study introduces an integrated probabilistic modeling approach for sequential crime risk assessment under the POCSO (Protection of Children from Sexual Offences) framework, combining the strengths of Hidden Markov Models (HMMs), stratified forensic analysis, and criminology-informed cyber-behavioural theory. Building on principles of digital stratigraphy, the model fuses heterogeneous evidence including digital logs, geospatial traces, criminological reports, and behavioural markers into temporally ordered risk states. Hierarchical Pattern Mining (HPM) is employed to extract recurring offender behaviour sequences, while Forensic Sequence Alignment (FSA) synchronizes multimodal evidence layers to reduce asynchronous inconsistencies and minimize false associations. Using the CSI-DS2025 dataset (25,000 multimodal stratified records), the system was evaluated through 10-fold cross-validation, Bayesian hyper parameter optimization, and structured train–validation–test splits. Results show that the HMM-based model achieves 91.8% accuracy, 92.4% precision, 89.7% recall, 90.9% F1-score, and a Stratigraphic Reconstruction Consistency (SRC) of 0.87, outperforming baseline forensic and statistical models. To contextualize risk prediction within broader cyber-offending patterns, the study also incorporates criminology theories routine activity theory, rational choice theory, and deterrence theory highlighting their relevance in modeling sequential child-related offenses. Findings demonstrate that integrating criminological logic with probabilistic modeling enhances offender-state transition interpretation, improves early-warning capabilities, and supports legally robust decision-making under POCSO guidelines. The proposed framework establishes a scalable foundation for real-time child safety analytics, forensic reconstruction, and cyber-crime prevention governance