Context-sensitive regression analysis for distributed data

Abstract

A precondition of existing ensemble-based distributed data mining techniques is the assumption that contributing data are identically and independently distributed. However, this assumption is not valid in many virtual organization contexts because contextual heterogeneity exists. Focusing on regression tasks, this paper proposes a context-based meta-learning technique for horizontally partitioned data with contextual heterogeneity. The predictive performance of our new approach and the state of the art techniques are evaluated and compared on both simulated and real-world data sets. © Springer-Verlag Berlin Heidelberg 2005.