Demand Forecasting for Fruit Chips: A Comparative Study of GRU and LSTM Models
Masud Effendi (*), Tiyo Ananda Rizky Bachtiar

Department of Agroindustrial Technology, Faculty of Agricultural Technology, Universitas Brawijaya
Veteran Street, Malang, East Java, Indonesia
*mas.ud[at]ub.ac.id

Abstract

The demand for fruit chip products at CV. Puri Pangan Lestari in Malang, Indonesia, shows a volatile, increasing trend, posing a challenge for effective inventory and production management. To address this fluctuation, this study aims to perform multivariate demand forecasting by comparing the accuracy of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. A multivariate forecasting approach was chosen to account for the interacting demands of jackfruit, snake fruit, and apple chips as influential variables. The data used for the research spans from May 2020 to December 2024. The study follows a series of steps, from data exploration and normalization to model compilation, training, and accuracy evaluation using MAPE metrics. The results consistently show that the multivariate GRU model outperforms the LSTM model. The optimal GRU model for jackfruit chip demand achieved an MAPE of 22.29%, while the best models for snake fruit and apple chips obtained MAPE values of 12.96% and 11.84%, respectively. The higher accuracy of the GRU model can be attributed to its simpler yet effective architecture. This research concludes that the multivariate GRU model is a more accurate and efficient approach for forecasting the demand for fruit chips at CV. Puri Pangan Lestari.

Keywords: Demand Forecasting- Fruit Chips- GRU- LSTM- Multivariate

Topic: Agro-industrial system management and regulation