Nonlinear Data Classification and Applications to Vehicle Calibration Identification

Abstract

In this paper, we consider the general problem of data classification, and focus on the development of suitable approaches for automotive calibration applications. Specifically, we propose two methods for nonlinear data classification: the first one is to find the smallest piecewise linear (PWL) region for a given data set and the second one is to construct piecewise quadratic (PWQ) boundary to separate two data sets . In both methods, the construction of the boundary curve is formulated as convex optimization problems that can be solved efficiently. Our approaches can incorporate prior information about data distribution and allow fixed structure of the decision boundary from different data sets with similar generating sources. We demonstrate the efficiency and effectiveness of the approaches with an application to calibration identification of vehicle rollover detection algorithm.