Comparative Analysis of Time Series Methods LSTM and ARIMA for Predicting Inventory Availability (Case Study: PT XYZ)

Abstract

Product availability plays a crucial role in supply chain management, directly impacting all aspects of business operations, from production to distribution. This study analyzes the optimization of product availability at PT. XYZ, a frozen and chilled food trading company in Indonesia, focusing on four main commodities: beef, buffalo meat, chicken, and potatoes. Utilizing historical transaction data from 2020 to July 27, 2024, this research compares the performance of two forecasting models: ARIMA (AutoRegressive Integrated Moving Average) and Long Short-Term Memory (LSTM), in predicting product availability The traditional ARIMA model has proven effective in time series data analysis but has limitations in capturing complex patterns and non-linear fluctuations. LSTM, as a machine learning technique, demonstrates superiority in capturing long-term temporal relationships. This study finds that the LSTM model consistently outperforms ARIMA for beef, buffalo meat, and chicken categories, although there is a slight increase in error for the potatoes category. Model performance evaluation is conducted using metrics such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (RMSE). The results indicate that the LSTM model exhibits lower errors compared to ARIMA, proving its effectiveness in predicting dynamic demand patterns. With a better understanding of product availability, the company is expected to reduce operational costs, avoid losses, and enhance customer satisfaction through more efficient supply chain management. This research provides significant insights for PT. XYZ and similar industries in implementing more accurate forecasting methodologies