Performance evaluation of evolving classifier algorithms in high dimensional spaces

Evolving systems and high dimensional stream data processing algorithms are of enormous practical importance and currently are under intensive investigation. This paper introduces an evolving neural classifier approach and evaluates its performance using high dimensional data and evolving and classic classifier algorithms representative of the current state of the art. The proposed approach works in one-pass mode to simultaneously find the neural network structure and its weights using high dimensional stream data. The results suggests that the classification rate achieved by the proposed approach is very competitive with the evolving models addressed in this paper. It outperforms all of them in most of the datasets considered. Also, the approach requires the lowest per sample processing time amongst the evolving and classic batch classifiers.