FM-based Supervised Learning for Categorical Data Classification in Manufacturing Process

For complex manufacturing process, the raw features are always in a heterogeneous manner that contains lots of high dimensional process parameters and categorical targets (e.g. product quality "Pass" or "Fail") without numerical measurements. It makes the classification problems quite challenging. To improve the classification accuracy, additional data structure is necessarily considered, e.g. feature interactions. Factorization Machines (FM) is a general classification method which considers the effect of feature interactions towards target. This paper discusses the classification performance of FM methods and its extensions to manufacturing processes and other industry fields. Real-world datasets are employed for the validation of classification performance. Experimental results show that FM-based approaches efficiently improve the classification accuracy about 2.49%-5.94% compared with conventional classification methods.