An Association Rule-Assisted Multi-Time-Series Forecasting method for non-production material consumption in the automotive sector

Non-Production Materials (NPMs) are a vital category of materials in automotive manufacturing, including various items whose consumption is often interconnected, especially with maintenance activities. Predicting NPM consumption is complex due to these interdependencies. Most existing research tends to focus on forecasting the consumption of individual materials, overlooking the advantages of utilizing data from multiple materials. This narrow focus limits both the accuracy and the breadth of forecasting efforts. In this paper, we formulate a multi-NPM consumption forecasting problem, which aims to predict the consumption of several NPMs at once. To address this issue, we introduce an Association Rule-Assisted Multi-Time-Series Forecasting Method (AR-MTSF). Our approach combines data from multiple materials with similar attributes and employs association rules to enhance forecasting accuracy. We assessed the effectiveness of AR-MTSF using a real-world NPM consumption dataset from a collaborating multinational automotive manufacturer. The experimental findings reveal that when forecasting automotive NPM consumption, the AR-MTSF method, when paired with the same forecasting algorithm, improves accuracy by 5%–30%.