A dynamic size artificial neural network for online data clustering with a new outlier handling technique

This paper presents a new online data clustering algorithm with a new outlier handling technique. The proposed algorithm procedure is based on the well-known ART networks. In recent years, ART networks have been widely used as an online data clustering technique in many applications. The problem with the ART networks is that when the network size increases due to the formation of new clusters, the clustering performance slows down. The situation will get worse if the incoming stream of data includes many outliers which will be processed by the network as new clusters. The proposed algorithm provides an online outlier handler which will solve the mentioned problem while categorizing the multi-dimensional input data using distribution-based clustering model. The outlier handling technique in the proposed algorithm could be used in other forms of ART networks such as ART1, ART2 and Fuzzy ART.