Time-Series Forecasting of School Abductions in Nigeria: An ARIMA-Based Early-Warning Framework

School abductions in Nigeria persistently undermine educational continuity and human security. While existing scholarship has examined the drivers and consequences of kidnapping, structured forecasting tools for proactive school protection are lacking. This study addresses that gap by developing a time series early warning framework using documented incident data from January 2013 to December 2021. Incident records from verified public domain sources were aggregated into a monthly time series and analyzed using an ARIMA (1,1,1) model in Python. Model identification was guided by autocorrelation diagnostics and by minimizing the Akaike Information Criterion. Forecast performance was evaluated using a temporally ordered 70/30 training–testing split with Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) as validation metrics. The model achieved a training RMSE of 34.99 and a testing RMSE of 96.81. The higher testing RMSE, while reflecting sensitivity to unforeseen escalation spikes, is acceptable for a probabilistic early warning tool where sensitivity is prioritized over exact precision. Findings reveal that school abductions exhibit structured temporal dependence and cyclical escalation–de escalation dynamics rather than random occurrence. Five-month-ahead forecasts generate probabilistic risk windows aligned with historical volatility patterns. Although not intended to predict specific attack events, the framework demonstrates that historical incident data can be transformed into actionable early warning signals for anticipatory school security governance. The study contributes to education sector security research by advancing a shift from reactive post-incident response toward predictive, data-informed protection strategies in conflict-sensitive environments.