Selecting Top-Ranked Solutions for Inverse Prediction With Multiple Responses Using JMP ®
Sep 7, 2017 1:15 PM
Lu Lu, PhD, Assistant Professor of Statistics, University of South Florida
Christine M. Anderson-Cook, PhD, Research Scientist, Los Alamos National Laboratory
Inverse prediction utilizes relationships between multiple inputs and outputs from a process to predict the most likely set of inputs (causes) that may have generated a set of newly observed outputs. It has important applications in contraband forensics and reverse detection problems. Current practice focuses on seeking a single best match for the inputs based on the observed responses by minimizing a measured distance between the new observation and the estimated responses at different locations. This talk presents an alternative approach, which uses Pareto front optimization to identify multiple leading solutions ranked with some quantification of robustness to different prioritizations of responses. A newly created add-in for JMP that implements the proposed methods will be demonstrated using a chemical process example.