A Hybrid Deep Learning and Time-Series Framework for Predictive Supply Chain Disruption Analytics: Design and Evaluation of SupplyChain Sentinel AI

The global supply chain ecosystem has evolved into an extraordinarily intricate network of interdependent logistics pathways, each susceptible to cascading disruptions arising from manufacturing delays, extreme weather events, geopolitical tensions, port congestion, and seasonal demand fluctuations. Despite the availability of historical shipment records, many organizations manage supply chain risk reactively-responding to disruptions only after they have already materialized and when mitigation options are severely constrained. This paper presents SupplyChain Sentinel AI, an intelligent, data-driven disruption-prediction system that combines deep learning, classical time-series forecasting, and multi-dimensional risk analytics into a unified analytical platform. The system is built on a synthetically generated yet operationally realistic dataset of 2,000 shipment records annotated with carrier identity, origin-destination city pairs, weather condition, traffic density, route distance, transit time goal, estimated and actual arrival timestamps, and the resulting delay in hours. Three forecasting approaches are implemented and rigorously benchmarked: an LSTM deep-learning network capturing non-linear multi-variable relationships; an ARIMA statistical model exploiting temporal autocorrelation in daily average delay; and a Facebook Prophet model decomposing the delay series into trend, seasonality, and residual components. The LSTM model achieves the lowest Mean Absolute Error of 21.81 hours, outperforming ARIMA (26.06 h) and Prophet (26.42 h). Four analytical visualizations-a route-level risk heatmap, a high-risk route ranking, a carrier reliability chart, and a weather-impact analysis-translate model outputs into actionable business intelligence, while a MySQL backend (InspireDB) provides persistent, queryable storage. The complete system is orchestrated through a modular five-stage pipeline executable as a single command, establishing a reproducible reference implementation for open-source, commodity-hardware supply chain risk management.