An Empirical Study and Application of Pharmaceutical Supply Chain Risk Perception Based on AI Prediction and Optimization Models

Abstract

The stability and reliability of medical supply chains are closely tied to both patient safety and the quality of healthcare services. In this paper, we address the issue of predicting supply disruption risks within medical supply chains. To do this, we develop a risk prediction model based on XGBoost and several time-series forecasting approaches. The model’s accuracy is further improved by incorporating dynamic factors such as inventory turnover rates and supplier ratings. We then conduct comparative experiments under four representative medical scenarios—including raw material shortages, sudden demand spikes, and logistics bottlenecks like port strikes—to assess the predictive performance of three time-series models: ARIMA, LSTM, and Transformer. The experimental findings indicate that in medical transportation scenarios, the LSTM model delivers the most accurate predictions of supply disruption risks. On the other hand, when it comes to emergency resource allocation, the Transformer model shows a clear advantage in issuing early warnings more promptly than conventional approaches, mainly because of its strength in capturing sudden and unexpected events. This study is not without limitations. For example, it does not take into account certain scenarios such as cold chain supply logistics. In addition, because of computational restrictions, the current models still face challenges in handling real-time, high-dimensional data. Looking ahead, future work could improve both accuracy and practical applicability by integrating IoT data or by refining hybrid model architectures.