An AI-Driven Demand Forecasting Framework for Agricultural Products to Reduce Food Waste and Price Volatility in Sri Lanka

Abstract

Agricultural production in Sri Lanka faces persistent challenges due to unreliable demand forecasting, leading to overproduction, food waste, and unstable market prices. This study develops an AI-driven demand forecasting framework to predict agricultural product demand and price fluctuations using machine learning and deep learning techniques. Key variables include sales volume, average price, weather conditions, time, and crop variety. Multiple models were implemented and evaluated, including Linear Regression, Random Forest, and Long Short-Term Memory (LSTM). Results show that the LSTM model achieved the highest accuracy (MAE = 12.4, RMSE = 18.7, R² = 0.89), outperforming Random Forest (MAE = 21.3, RMSE = 29.5, R² = 0.74) and Linear Regression (MAE = 34.1, RMSE = 42.8, R² = 0.52). These findings highlight the importance of temporal dependency modelling in agricultural forecasting. The proposed framework integrates a decision-support component to assist farmers in production planning, thereby reducing food waste and minimizing price volatility. The study concludes that AI-based forecasting can contribute to sustainable agricultural practices and improved economic stability in Sri Lanka's farming sector.