A latent variable clustering method for wireless sensor networks

In this paper we derive a clustering method based on the Hidden Conditional Random Field (HCRF) model in order to maximizes the performance of a wireless sensor. Our novel approach to clustering in this paper is in the application of an index invariant graph that we defined in a previous work and that precisely links a hyper-tree structure to the data set assumptions. We show that a set of conditional index invariant hyper graph forms a tree and then, we show that any tree factorization optimizes the conditional probability of an HCRF model. We evaluate our method based on a custom data set that we obtain by running simulations of a time dynamic sensor network. The performance of the proposed method outperforms the existing clustering methods, such as the Girvan-Newmans algorithm, the Kargers algorithm and the Spectral Clustering method, in terms of packet ac ceptance probability and delay.