Pattern Classification Based on Local Learning

Local learning methods approximate a target function (a posteriori probability) by partitioning the input space into a set of local regions, and modeling a simple input-output relationship in each one. In order for local learning to be effective for pattern classification in high dimensional settings, regions must be chosen judiciously to minimize bias. This paper presents a novel region partitioning criterion that attempts to minimize bias by capturing differential relevance in input variables in an efficient way. The efficacy of the method is validated using a variety of real and simulated data.