As the fabrication process of integrated circuit substrates (IC substrates) becomes more complicated, higher capital-intensity equipment is adopted, making equipment utilization critical to maintaining a competitive advantage. There has been little research into machine learning applications for the IC substrate build-up process. This study aims to develop a data-driven framework to predict and monitor equipment health by integrating extreme gradient boosting regression with exponentially weighted moving-average methods, based on status variable identification (SVID) analysis, and providing maintenance recommendations according to domain knowledge based on different health statuses. The proposed method has been validated in a leading IC substrates company in Mainland China. Results have shown that the proposed approach can provide early detection of abnormal equipment health status, reducing unplanned downtime, capacity loss, and direct labor costs for repairs as part of efforts toward Industry 3.5.