Tuning the Covariate Influence in a Gamma Process Clustering Algorithm

Edith Grall-Maës (1) , Xuanzhou Wang (1) , Pierre Beauseroy (1)

1 LM2S - Laboratoire Modélisation et Sûreté des Systèmes

This paper addresses the problem of clustering stochastic deterioration processes in a Gamma process model-based framework. The process evolution is assumed to be described by realizations of Gamma processes. Complementary information is given by covariates which characterize the systems from which the realizations originate. The processes parameters depend on a partition of the covariate space. To identify the process parameters one need to cluster data taking into account the realizations and the covariates. The proposed method embeds both concerns by using an EM algorithm with side-information and a local a posteriori probability. The weight given to the Gamma process realizations with respect to the covariates depends on the importance of the neighborhood considered in the local a posteriori probability. Criteria related to the value of the parameter tuning this importance are proposed. Simulated data is used to illustrate the clustering method and to study the influence of the parameter tuning the covariate influence on the error result and on the proposed criteria.

Domains: Chemical Sciences

Edith Grall-Maës, Xuanzhou Wang, Pierre Beauseroy. Tuning the Covariate Influence in a Gamma Process Clustering Algorithm. Chemical Engineering Transactions, 2013, 33, pp.49-54. ⟨10.3303/CET1333009⟩. ⟨hal-02365809⟩