Kombinasi Model ARIMA dan KNN Dalam Peramalan Harga Produk

Abstract

This study proposes a product price forecasting model for PT ABC by integrating the Autoregressive Integrated Moving Average (ARIMA) model and the K-Nearest Neighbor (KNN) method into a hybrid predictive approach. The company faces recurring challenges related to product price fluctuations and stock availability caused by unstable market conditions and irregular supply distribution. To address these issues, a data-driven forecasting model is required to support inventory planning and price stabilization strategies. The dataset used in this study consists of historical cement purchase records from January 2023 to September 2025, obtained from the company’s ERP system. The research process includes data cleansing, transformation, monthly price aggregation, and the application of ARIMA, KNN, and a hybrid ARIMA–KNN model designed to improve forecasting accuracy. The evaluation results indicate that the hybrid ARIMA–KNN model outperforms the standalone ARIMA model in short-term price forecasting. Based on three performance metrics, the hybrid model achieved a Mean Absolute Error (MAE) of 1604.94, a Root Mean Square Error (RMSE) of 2299.37, and a Coefficient of Determination (R²) of 0.2881. These results suggest that while the model captures a portion of price variability, it still faces limitations in modeling non-linear fluctuations and sudden extreme changes. Nevertheless, the hybrid approach demonstrates improved stability by reducing extreme prediction variations, maintaining trend continuity, and generating smoother prediction curves that more closely align with actual price movements. This research contributes practically by providing PT ABC with a forecasting tool to support future price estimation, improve inventory management, and maintain market price stability. Additionally, the findings offer a foundation for future research on advanced non-linear and deep learning–based forecasting models.