Backpropagation Neural Network Model for Predicting Spare Parts Demand Under Dynamic Industrial Conditions

Purpose – This study aims to analyze and control spare parts inventory in pumping units at PT. XYZ using the Artificial Neural Network (ANN) method. The research addresses the challenges of surplus and shortage of spare parts, which directly affect operational continuity, production costs, and company performance. Design– A qualitative approach combined with quantitative modeling was employed. Data were collected through observation, the dataset was normalized and divided into three training-testing scenarios (70:30, 80:20, and 90:10). The ANN model with backpropagation was developed and tested using Matlab software, with accuracy evaluated through Mean Squared Error (MSE) and correlation coefficient (R). Findings – The results show that Scenario 2 (80% training and 20% testing data) provides the best balance, yielding the highest accuracy. The ANN model captured nonlinear inventory patterns, achieving very low MSE (3.1358e-12) and demonstrating predictive reliability. However, the overall correlation (R = 0.6015) indicates the need for larger datasets and model refinement to improve generalization. Practical implications – Applying ANN in inventory management helps companies minimize risks of overstock and shortages, reduce storage costs, and support reliable production planning. This contributes to supply chain resilience and enhances customer trust in operational performance. Originality – This study presents one of the first applications of ANN for spare parts inventory prediction in Indonesia’s pumping unit sector. The findings provide empirical evidence of ANN’s effectiveness and offer theoretical as well as practical contributions to the advancement of AI-based inventory management in industrial contexts.