Improving the Accuracy of Food Commodity Price Prediction Model Using Deep Learning Algorithm

Abstract

The world market for agricultural commodities is essential to maintaining both economic stability and food security. However, due to its intrinsic volatility, this market is subject to price fluctuations caused by a variety of variables, including supply chain interruptions, geopolitical events, and economic conditions. Predicting food commodity prices accurately and on time is essential for all parties involved, including farmers, traders, policymakers, and consumers. The existing method proposed a hybrid LSTM-CNN model to forecast weekly prices of oats, corn, soybeans, and wheat in the U.S., finding that hyperparameter tweaking over 15 weeks affected its accuracy. Despite its strengths, the LSTM- CNN model faced challenges such as complexity, computational cost, and overfitting, highlighting the need for better optimization and hybrid approaches to improve prediction accuracy. The Whale Optimization Algorithm (WOA) was used in this study to optimize hyperparameters and train deep neural network architecture for food commodity price prediction in Nigeria. The study utilized four performance metrics: RMSE, MSE, MAE, and R2. The proposed model achieved the lowest RMSE (0.0071-0.0073), MSE (0.0061), and MAE (0.0082-0.0083) values, indicating higher accuracy in predictions compared to CNN-LSTM and CNN models. Additionally, it achieved the highest R2 values (0.972-0.975), further demonstrating its superior performance in forecasting food commodity prices.