Agroindustry Supply Chain Optimization Using Machine Learning Based Demand Prediction and Genetic Algorithms

Demand uncertainty and the perishable nature of products present major challenges in agro-industrial supply chain management, frequently leading to inflated operational costs due to stock accumulation or missed sales opportunities. This study proposes a hybrid approach that integrates deep learning and meta-heuristic methods to address these issues. Specifically, a Long Short-Term Memory (LSTM) model is employed to forecast stochastic demand, the outputs of which are subsequently fed into a Genetic Algorithm (GA) to determine optimal order quantities. A case study was conducted using sales data for red chili commodities. The empirical results demonstrate that the LSTM model predicts demand with high precision, yielding a Mean Absolute Percentage Error (MAPE) of 8.5%, which significantly outperforms the conventional ARIMA method. Furthermore, integrating these predictions into the GA optimization model successfully minimizes the total supply chain cost by 25.0% compared to the traditional Economic Order Quantity (EOQ) approach. The most substantial cost reduction was observed in shortage costs, which decreased by 84.0%, indicating the model's efficacy in preventing stockout events. This research contributes an adaptive decision-making framework to enhance logistics efficiency in the agro-industry.