Design of Industrial Machine Learning System

A large number of real-world machine learning and data mining problems have several crucial issues such as class imbalance and multiple objectives to achieve, which cannot be coped with many learning methods easily.  This study proposes an efficient approach to address those issues and applies the approach in design of machine learning system for a real-world problem − inline inspection of surface defects on glass substrates of thin-film transistor liquid crystal display (TFT-LCD). Two major steps of decision support in machine learning system are: (1) use of threshold-moving to handle data imbalance and achieve individual quality objectives, and (2) utilization of Ensemble techniques to handle multiple quality objectives in a classification problem while maintaining low computational complexity. When applied to the industrial case study, the achieved performance shows that utilization of the proposed approach in defect inspection of TFT-LCD glass substrates would be a viable alternative to manual inspections.