Interpretable Query Projection Learning

Abstract

We propose a novel lazy learning method, called QPL, which attempts to discover useful patterns from training data. The discovered patterns (Query Projections) are customized to the query instance and easily interpretable. As a pattern discovering method, QPL does not require a batch training process and still achieves excellent classification quality. We use some benchmark data sets to evaluate QPL and demonstrate that QPL has a prominent performance and high reliability. © Springer-Verlag 2004.