On applying the Restricted Boltzmann Machine to active concept drift detection

In this paper the issue of active concept drift detection in time-varying data stream mining is taken under consideration. The Restricted Boltzmann Machine (RBM) is proposed to be applied as a drift detector. The RBMs are neural networks which are able to learn generative models of data. After training they contain a compressed information about the distribution from which the training data were drawn. We suppose that such an RBM learned on a part of the data stream can then be used to determine whether the new data elements from the stream are drawn from the same distribution or not. Two indicators are proposed to be used for evaluation of incoming data: the free energy and the reconstruction error. Their computation is relatively fast, hence they are suitable for data stream scenario. Preliminary experimental results demonstrate that the proposed tool is able to deal with different types of concept drift, e.g. the sudden or the gradual.