Collaborative Research: Variable Selection for Remedying the Effects of Uncontrolled Variation in Data Driven Predictions

点击翻译按钮获取中文摘要

In a project funded by the Chemical Measurement and Imaging Program of the Chemistry Division, Professors Steven Brown and Karl Booksh, both from the University of Delaware, and Professor Barry Lavine of Oklahoma State University, seek to improve mathematical and statistical modeling of real-time measurement of chemical species. This research is aimed at reducing the cost-of-ownership of calibration and classification models that have become increasingly common in process monitoring, quality assurance and quality-by-design applications by studying the ways in which these models fail and by searching for ways to stabilize the modeling. The broader impacts of this project include the improvements in efficiency and performance from improved and more reliable chemical models for production and quality assurance, as well as the training of graduate and undergraduate students in data analysis, a skill in very high demand.This project is a collaborative effort aimed at investigating fundamental issues important to chemical modeling in modern measurement science: (1) stabilizing calibration and classification relationships in the presence of unexpected contributions, (2) making calibrations and classifications as robust as possible to additional sources of background, and (3) improving classification and calibration models measured over multiple instruments. Methods for automated identification of the portions of the chemical responses that best model the system are being investigated.