Selection of Relevant Parameters Based on Conflict

Abstract

In this paper, we present a new approach for selection of relevant parameters in fault detection process. The conflict notion is used in order to evaluate the relevance of any subset of parameters. This selection is divided in two steps. The first consists in making partitioning of data running smooth. During the second step, we calculate the conflict between each subset of parameters. The experimental results are presented for several subsets of parameters for given breakdown in automobile system. In the second part of paper, we propose an automatic discretization algorithm; this algorithm improves the Hellinger-based discretization by adding an automatic stopped criterion.