eGauss+ evolving clustering in classification

In this paper, eGauss+ evolving clustering is used in classification, which forms very small clusters in the shape of hyper-ellipsoids in a single-pass manner going through the data set. At the end of procedure small cluster forming phase the merging procedure is used which merges these small clusters, i.e. merges granules into bigger clusters. The merging procedure is based on cluster volumes, which merge two closed and similar cluster into a new one. The resulting cluster center is a weighted averaging of merged cluster centers, together with covariance matrix which is calculated from the covariance matrices of the small clusters. The proposed classification algorithm was used on two classical classification data sets, i.e. iris and breast cancer data set, and compared with other methods. The method shows very similar results, but has an important advantage, namely it works recursively, in a single-pass manner.