The Multivariate Forecasting of Chicken and Beef Prices Involving Weather, Economic, and Health Factors Using the Gated Recurrent Unit Method

Abstract

Food security, especially in the livestock sector in the form of broiler chicken and beef cattle, is a strategic issue for Indonesia to always be able to balance supply and demand for these food commodities. Food price forecasting is needed to mitigate rising food prices for these commodities. Previous research on food price forecasting was only univariate forecasting and comparison of error results between forecasting algorithms. This study aims to perform multivariate forecasting of broiler and beef cattle prices in DKI Jakarta by involving weather, economic, and health factors using the Gated Recurrent Unit (GRU) algorithm where the accuracy test is based on the MAPE value. The GRU algorithm for multivariate forecasting of broiler and beef cattle prices yielded an average MAPE for training and testing of 0.471% and 1.150% indicating that all models in the very good accuracy category for multivariate forecasting of broiler and beef cattle were represented. In addition, the model also produces deviations between MAPE values in the training data and test data which are not too different so that the model developed with each price of broiler chicken and beef cattle is categorized in the best fitting category.