A New Just-In-Time Modeling Method and Its Applications to Rolling Set-up Modeling

Abstract

This paper proposes a variant of k Bipartite Neighbors (k-BN), called k-BN2, for function prediction. Like k-BN, k-BN2 selects k instances surrounding a query, i.e., a novel instance, and keeps them bipartitely. However, in order to improve the prediction precision, based on the bipartite neighborhoods, k-BN2 combines local linear models and global nonlinear model to predict the value of the novel instance. Applied to two real measured datasets, k-BN2 outperforms typical k-BN and those methods in which k-BN or related approximate physical model alone is used.