Knowledge discovery in data streams with the orthogonal series-based generalized regression neural networks

In this paper, a method for nonparametric regression estimation in a time-varying environment is presented. The orthogonal series-based kernels are used to design learning procedures tracking non-stationary systems changes under non-stationary noise. The presented procedures, constructed in the spirit of generalized regression neural networks, are a very effective tool to deal with stream data. The convergences in probability and with probability one are proved, experimental results are given and discussed.