Hybrid GARMA-LSTM Model for Enhanced Demand Forecasting and Optimal Resource Distribution

Modern supply chains rely on effective demand forecasting and allocation of resources. Traditionalised models, such as the Generalized Autoregressive Moving Average (GARMA), are proficient in the prediction of linear temporal dependencies but are generally incapable of managing the non-linear complexities of real-world demand patterns. In this study, we propose a new hybrid model by integrating GARMA and Long Short-Term Memory neural networks (LSTM) for improving demand forecasting and resource allocation.

GARMA is used for linear trends, while LSTM can capture more complex, non-linear behaviours. By doing so, the model can better analyze the demand in such situations of change or fluctuations, and can gain valuable insight on market demand for better allocation of resources. The hybrid model exceeded traditional supply chain methods during multiple testing scenarios to provide substantial improvement. The model achieved forecast accuracy improvements up to 20% which subsequently optimized both inventory control and decreased expenses while improving service quality. The proposed model generates optimized frameworks for forecasting supply chain demand alongside resource distribution operations.