Key points are not available for this paper at this time.
The direct impact of late deliveries on supply-chain performance is critical, as it directly affects customer satisfaction and business income. Most conventional machine-learning-based risk-predictive models do not allow for transparency regarding determining supplier performance or identifying the reasons behind delivery delays. This research thus attempts to fill in the gaps of the models above by proposing an integrated analytical framework for real-time supplier evaluation and disruption forecasting. The hybrid model proposed utilizes Data Envelopment Analysis and Temporal Convolutional Networks for supplier-efficiency assessment with late-delivery risk prediction. The DEA model, particularly the Banker-Charnes-Cooper model, finds efficiency scores based on inputs such as lead time and costs per shipment and outputs such as on-time delivery rate and sales per customer. These efficiency scores are further fed, along with temporal features, to the TCN model, which captures historical patterns with dilated causal convolutions and residual connections to predict the probability of incidence of late deliveries. The integrated DEA-TCN model surpasses conventional approaches, with 99 % accuracy, 99 % precision, and 98 % recall. It serves to distinguish inefficient suppliers as well as high-risk transactions for making pre-emptive decision-making. Experimental evidence substantiates the viability of forecasting disruption and improving supply chain resilience. This transparent, data-driven framework gives managers actionable insights to optimize supplier selection, diminish operational risks, and improve delivery reliability. The matrix with a DEA-TCN foresees the strategic planning and strengthens the adaptability of supply chains in highly dynamic environments.
Abushaega et al. (Wed,) studied this question.