Selection and Comparison of Supervised Predictive Data Mining Models for Electronics Fabrication Data

Abstract

In order to predict the performance of a manufacturing process or system, proper mathematical models are needed. This research investigates the use of two competitive unsupervised data mining methods - regression and neural networks - in developing an empirical model for two electronics fabrication processes/systems. A case study from experimental data of electronics fabrication is used to demonstrate how to deal with these issues when regression and neural networks models are used for the purpose of prediction. It will be shown that hypothesis tests and cross-validation are valuable in validation, selection and comparison of predictive models. A rigorous procedure is proposed for construction, validation, selection, and comparison of regression and neural networks models applied to predictive modeling of experimental data.