FOOD DEMAND PREDICTION USING NON LINEAR AUTO REGRESSIVE EXOGENOUS NEURAL NETWORK

Abstract

Accurate food demand prediction is a critical challenge in modern supply chain management, particularly in sectors such as restaurants, catering services, and food delivery platforms where demand patterns are highly dynamic and influenced by multiple external factors. This project proposes a robust predictive framework based on a Nonlinear Autoregressive Exogenous (NARX) Neural Network, designed to model complex temporal dependencies and nonlinear relationships in food demand data. The NARX model leverages both historical demand values and exogenous inputs such as weather conditions, promotional activities, seasonal trends, holidays, and customer behavior patterns to generate precise forecasts. The proposed system employs a supervised learning approach where time-series data is preprocessed and structured into input-output sequences suitable for NARX training. Advanced feature engineering techniques are applied to capture hidden patterns and correlations among variables. The neural network architecture is optimized through hyperparameter tuning to enhance prediction accuracy and generalization capability. Performance evaluation is conducted using metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R²), ensuring a comprehensive assessment of model efficiency. Experimental results demonstrate that the NARX-based model significantly outperforms traditional statistical methods and basic machine learning models in capturing nonlinear demand fluctuations. The system enables businesses to optimize inventory management, reduce food wastage, and improve customer satisfaction through timely availability of resources. This research highlights the potential of advanced neural network architectures in addressing real-world forecasting problems and contributes to the development of intelligent, data-driven food supply systems.