Banditry and National Development in Nigeria: A Spatio-Temporal Machine Learning Approach to Conflict Prediction

Banditry has emerged as one of the most severe internal security challenges confronting Nigeria, with far-reaching implications for national development, governance, and social stability. This study applies spatio-temporal machine learning techniques to analyse and predict patterns of banditry incidents across Nigerian Local Government Areas (LGAs) between 2015 and 2024. Drawing on Routine Activity Theory, Social Disorganisation Theory, and Government–Development Theory, the research integrates socio-economic, environmental, and security-related variables with historical conflict data comprising over 20,000 recorded incidents. A comparative modelling framework using Random Forest (RF), Long ShortTerm Memory (LSTM), and Convolutional LSTM (ConvLSTM) models is implemented to capture non-linear relationships, temporal dependencies, and spatial diffusion of violence. Results indicate that ConvLSTM outperforms other models in forecasting high-risk locations and attack severity, demonstrating the value of Spatio-temporal deep learning for conflict prediction. The findings reveal persistent hotspots in areas characterised by weak state presence, high poverty levels, arms proliferation, and recurrent attack cycles. The study concludes that machine-learning-driven early-warning systems can enhance proactive security planning, optimised resource allocation, and mitigate the developmental impacts of banditry in Nigeria.