Inductive Learning Using Multiscale Classification

Multiscale Classi cation is a simple rule-based inductive learning algorithm. It can be applied to any N -dimensional real or binary classi cation problem to successively split the feature space in half to correctly classify the training data. The algorithm has several advantages over existing rulebased and neural network approaches: it is very simple, it learns very quickly, there is no network architecture to determine, there is an associated con dence with each classi cation rule, and noise can be automatically added to the training data to improve generalization.