Regional Crop Yield Prediction Using Feedforward Neural Networks

Crop Yield Prediction or CYP is a crucial aspect of food security. Being able to forecast shortages and excess in crop production can aid consumers in supply management as well as help the government in creating agricultural policies and implementing economic decisions that can help balance the supply chain. This study aims to build a crop yield prediction model using Neural Networks that is capable of handling high-dimensional datasets with complex patterns. The model was trained using crop yield data from the province of Quezon for years 1987-2015, as well as the climatological data observed in the same province for the same period. The model was tested using the same data features that were observed for the years 2016-2022. The forecast performance of the resulting model was benchmarked against two statistical forecasting techniques, namely the univariate Seasonal Autoregressive Integrated Moving Average and the multivariate Autoregressive Distributed Lag using the performance metrics Mean Absolute Error, Root Mean Squared Error, and Mean Absolute Percentage Error. The research findings indicate that the Neural Network-based CYP model performed best for both rice and corn yield forecasting with a MAPE of 3.84% and 4.46%, respectively.