Modeling concept drift from the perspective of classifiers

The problem of concept drift is of increasing importance to data mining as more and more data is organized in the form of data streams rather than static database, and it is rather unusual that concepts and data distribution stay stable over time. In this paper, we model the concept drift as the changes of the optimal parameters of the discriminative model. By employing the extended Kalman filter to track the optimal parameters in this model, we get a dynamical classifier with adaptability to the dynamics of concept drift. The empirical results in both synthetic and real data sets indicate that the proposed algorithm is an effective and efficient solution to classification for evolving data streams.