Forecasting of Rice Crop Production Using a Recurrent Neural Network

Abstract

Rice holds immense significance in Filipinos' diets, is consumed nearly three times daily, and is a crucial dietary component. Using a recurrent neural network (RNN), this study seeks to develop and evaluate a web application for predicting rice crop production in Davao del Sur. The system enables the Davao del Sur agriculturist to anticipate rice crop production over the next four quarters. The model used in the system was designed to process and analyze sequential data, such as time series, by capturing and retaining long-term dependencies and patterns. The researcher used the rapid application development model for system development. It started with data preprocessing and prototyping, which involved close coordination with end users to collect their feedback for interface enhancements. The model's performance is evaluated using validation loss, training loss, mean squared error (MSE), and R-squared metrics. The model's parameters were fine-tuned by utilizing these metrics, resulting in an improved capability to learn from data and generate accurate forecasts. The web application offers administrators a login feature, providing them with data administration and reporting functionalities. In addition to accessing the forecasting information, users can generate reports to assist with decision-making. The model is trained and validated in testing, and the front-end's functionality is evaluated. The training results indicated the optimal model configuration; the optimal epoch for the model with a batch size of 16 is 200.