<b>A CRIME PREDICTION MODEL USING DEEP LEARNING BERT FOR LAW ENFORCEMENT AGENCIES IN NIGERIA</b>

This study aims to develop a crime prediction model that combines Deep Learning, BERT, LSTM, and GCN to support proactive policing and enhance public safety. The proposed framework employs BERT to analyze contextual data in crime-related text, LSTM to capture temporal dynamics, and GCN to model relationships between crime types, locations, and entities. Datasets were collected from ngCERT (2019–2025), the Nigerian Police Force, and open-source reports from Vanguard and Premium Times. The system's performance was evaluated using accuracy, precision, recall, F1-score, and ROC metrics. The system achieved 97.3% accuracy, which is 10.3% higher than the base research models considered. The results indicate that the hybrid transformer-based model achieves higher performance than conventional Machine Learning methods, such as Random Forest, Support Vector Machine, K- Nearest Neighbors, and CART, for crime prediction. The findings highlight the capability of Deep Learning transformers to improve crime forecasting and offer a practical framework for enhancing resource allocation and decision-making within Nigerian law enforcement agencies. The model can also be integrated into a security information and event management platform to give real-time predictions.