Deep Learning-Enhanced Comparative Analysis of ARIMA, Seasonal ARIMA, and Gated Recurrent Unit Models for Forecasting Car Sales

Fariz Zakaria (Faculty of Informatics Engineering, Amikom University, Indonesia)
Ema Utami (Faculty of Informatics Engineering, Amikom University, Indonesia)

International Journal of Informatics, Information System and Computer Engineering (INJIISCOM)
Vol. 7 No. 1 (2026): INJIISCOM: VOLUME 7, ISSUE 1, JUNE 2026 (ONLINE FIRST)

This study aims to assess and compare the performance of three forecasting models—Autoregressive Integrated Moving Average (ARIMA), Seasonal Autoregressive Integrated Moving Average (SARIMA), and Gated Recurrent Unit (GRU)—in predicting Toyota car sales in Indonesia. The dataset exhibits intricate seasonal patterns and non-linear fluctuations, posing significant challenges for conventional statistical approaches. The ARIMA model, typically suited for linear and stationary data, struggled to capture the complexities of these sales trends. Similarly, despite being designed to handle seasonal components, the SARIMA model failed to provide accurate predictions and demonstrated high error rates. In contrast, the deep learning-based GRU model exhibited superior predictive accuracy, achieving significantly lower values across all evaluation metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Ultimately, the findings confirm that the GRU model outperforms traditional statistical methods, offering highly accurate and reliable forecasts to support complex business decision-making in the automotive industry.