Boosting the Hierarchical Hyperellipsoidal Neural Gas Networks

This paper proposes a new classification scheme by using the hyperellipsoidal neural gas (ENG) networks and boosting methods. The presented ENG network inherits all the advantages of traditional neural gas networks especially with better adaption to gaussian-based distribution datasets for clustering analysis comparing to Kohonen´s self-organizing map and K-means etc. The soft competitive learning of ENG is based on local principal subspace, which can be applied to solve pattern recognition problem. In order to improve the classification ability of hierarchical ENGs, boosting methods are implemented for better finaldecision by using weighted sampling approach. The proposed scheme is used to the domain of intrusion detection. Some experiments are carried out on the KDD CUP 1999 Intrusion Detection Evaluation dataset.