COMPARATIVE ANALYSIS OF ARIMA AND SUPPORT VECTOR REGRESSION FOR DEMAND FORECASTING AND SPARE PARTS MANAGEMENT IN COLD FLEET REPAIR AND MAINTENANCE
Abstract
This study presents a comparative analysis of ARIMA (Auto Regressive Integrated Moving Average) and Support Vector Regression (SVR) methodologies for demand forecasting and spare parts management in the context of cold fleet repair and maintenance operations. The efficacy of ARIMA modelling investigated through a comprehensive examination of spare parts demand at Hutama, a leading service provider in Indonesia, spanning from January 2019 to December 2023. Utilizing time series plotting, stationarity tests, and model selection criteria, ARIMA (0, 0, 1) is identified as the optimal forecasting model, providing valuable insights into projected spare parts usage. Concurrently, the study explores the effectiveness of SVR in accurately predicting spare part usage within the service and maintenance industry, focusing on coolers and freezers. Leveraging historical data from Hutama, the SVR model achieves remarkable precision through rigorous parameter tuning. Comparative analysis reveals differences in forecasting accuracy, with ARIMA demonstrating a MAPE (Mean Absolute Percentage Error) of 15.79% and RMSE (Root Mean Square Error) of 15.45, while SVR exhibits a MAPE of 10.10% and RMSE of 9.35. This comparative analysis contributes to enhancing spare parts management practices and operational efficiency in the service and maintenance industries.