Application of Backpropagation Artificial Neural Networks for Optimizing Corn Production Prediction in Karo Regency

Abstract

Corn production in Karo Regency, North Sumatra, plays a crucial role in supporting regional food security and the local economy. However, fluctuations in production caused by unpredictable environmental conditions and limited data-driven forecasting methods have made it difficult for policymakers and farmers to plan effectively. This study aims to address this problem by developing a model to predict corn production using the Backpropagation Neural Network (BPNN) method. The study utilized 302 cleaned datasets, with Planted Area and Harvested Area as input variables, and Production as the output variable. The dataset was divided into 70% for training and 30% for testing. Five BPNN architectures (ranging from 2-4-1 to 2-12-1) were tested using three activation functions (Sigmoid, ReLU, and Tanh), with a maximum of 200 iterations and a learning rate of 0.01. The best results were achieved by the 2-12-1 architecture with the Tanh activation function, obtaining an R-squared value of 94.86% and a Mean Squared Error (MSE) of 0.0039. These findings demonstrate that the Backpropagation Neural Network is effective for forecasting corn production and can serve as a valuable decision-support tool for sustainable agricultural planning in the region.