ABSTRACT
This paper addresses the problem of deterioration process classification with the effect of covariates. A deterioration process is usually defined as a time-dependent stochastic process characterized by some unknown parameter vector. An evolution of the process is considered as a path that consists of several degrading measures which characterize the attribute values. It can be considered that different paths with the same parameter vector arise from the same deterioration process. Also, each path is supposed to be associated with a covariate value on which the parameter vectors depend, and we assume that nearby paths in covariate space tend to belong to the same process while distant ones may not. Thus, possessing a set of observations composedof attribute and covariate values, we aim to group the similar observations by taking into account the proximity in both attribute space and covariate space. The main difficulty lies in combining appropriately these two proximities, as they often represent different types of information in different scales. Based on such context, we propose a method that balance automatically both proximities in the process framework. The effectiveness of this method is validated through the experimental results by using simulated reliability data set which follows the Gamma process.
