Robust Supply Chain Forecasting Through Multi-Head Attention and Levy Flight-Based Optimization

Supply Chain Management enables customer satisfaction to the customers by enabling effective supply chain forecasting, resulting in low costs and greater logistical efficiency. The performance obtained by the conventional strategies is not effective due to the ineffective pattern learning in supply chain data. In particular, they have trouble dynamically prioritizing influential features and modeling long-term temporal dependencies. To resolve these issues, the Levy Flight Binary Dragonfly Optimization strategy with a Multi-Head Swift recursive neural network is proposed. The proposed model assists in achieving non-sequential memory transitions by introducing the Frame-Hopping Transmission Gate unit. This model allows the proposed model to be effectively used for long-term dependencies. To understand the time-series data, the Multi-Head Attention model was deployed in the developed approach, contributing to greater forecasting accuracy. The hyperparameter tuning of this study ensures that the proposed model reaches better performance values by applying Levy Flight Binary Dragonfly Optimization. Among other forecasting methods, thorough experimental analyses are carried out to guarantee the proposed model’s capability. The developed model exhibits a reduced Mean Absolute Error of 1.5 and greater accuracy of 98.97%. In the experimental analysis, the effectiveness of each element of the proposed forecasting model is demonstrated by the ablation analysis.