Feature Selection and Negative Binomial Regression for Predicting Number of Defects in Wire Mesh Production

• Main Authors: Rungchat Chompu-Inwai, Aree Wiriyaphongsanon, Trasapong Thaiupathump
• Format: Conference Proceeding
• Published: 2019
• Subjects: Business, Management and Accounting; Decision Sciences; Engineering
• Online Access: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85066608371&origin=inward; http://cmuir.cmu.ac.th/jspui/handle/6653943832/65429
• Institution: Chiang Mai University

© 2019 IEEE. In wire mesh production, many types of defects are found. When the factors related to the number of defects occurring are correctly identified, various improvement methods can then be applied to reduce or control the number of defects. In this paper, the features that are strongly linked with the number of defects are identified by the feature selection process and then used in the prediction process. LASSO method and random forest are applied in the feature selection process. Using selected features from feature selection, a negative binomial generalized linear model (GLM) is employed to predict the number of defects in the mesh manufacturing process. A negative binomial regression is used since the nature of the mesh defect data in this study is count data and over-dispersed. Quality of the selected features from LASSO and random forest are compared using RMSE and RMSLE of the predicted results from the negative binomial regression.