On the Complexity of Rule Discovery from Distributed Data

This paper analyses the complexity of rule selection for supervised learning in distributed scenarios. The selection of rules is usually guided by a utility measure such as predictive accuracy or weighted relative accuracy. Other examples are support and confidence, known from association rule mining. A common strategy to tackle rule selection from distributed data is to evaluate rules locally on each dataset. While this works well for homogeneously distributed data, this work proves limitations of this strategy if distributions are allowed to deviate. To identify those subsets for which local and global distributions deviate may be regarded as an interesting learning task of its own, explicitly taking the locality of data into account. This task can be shown to be basically as complex as discovering the globally best rules from local data. Based on the theoretical results some guidelines for algorithm design are derived.