Battery Sales Forecasting Using ARIMA at Store X for Monthly Data Based on Battery Brand

Abstract

Demand uncertainty poses a persistent challenge for sales planning and inventory management in retail businesses, particularly when products exhibit heterogeneous sales patterns across brands. This study aims to develop brand-specific vehicle battery sales forecasting models using the Autoregressive Integrated Moving Average (ARIMA) method. The study follows the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework, comprising Business Understanding, Data Understanding, Data Preparation, Modeling, and Evaluation. Historical sales transaction data from Toko Aki Restu covering July 2016 to July 2026 were aggregated into monthly sales series for nine battery brands and divided into training and testing sets using an 80:20 ratio. Model development involved stationarity assessment using the Augmented Dickey-Fuller test, preliminary model identification through the Autocorrelation Function and Partial Autocorrelation Function, and parameter combination selection using Grid Search. Forecasting performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The findings indicate that the optimal ARIMA specification varies across battery brands, reflecting differences in their underlying sales patterns. GS ASTRA MF achieved the lowest MAE and RMSE, whereas AMARON recorded the lowest MAPE among the evaluated brands. The selected models were subsequently used to forecast monthly sales for a 12-month period from August 2026 to July 2027. These findings demonstrate that brand-specific ARIMA modeling can accommodate heterogeneous sales characteristics and provide differentiated forecasts for vehicle battery products. The study contributes a systematic forecasting approach that can support more informed sales planning and inventory management by aligning forecasting models with the time-series characteristics of individual brands.