Constraint excluded classifier
H. Abbassi, Reza Monsefi, Hadi Sadoghi Yazdi · 2012
Linear classifiers have the generalization property while lacking the power of classifying complex patterns. A simple and effective idea is to somehow exclude the complexity of the data such that it can be classified using a linear classifier. In this paper a new classifier system called “Constraint Excluded Classifier” is proposed that classifies most of the input patterns using a simple, e.g., a linear classifier. The classification is composed of an iterative three step loop. In the “Construction” step, several sub-classifiers are constructed which are responsible for linearly classifying parts of input patterns. Sub-classifiers are merged together in the “Fusion” step. The “Evaluation” step tests and fine tunes the construction of sub-classifiers. The comparison of the new classifier with famous classifiers is also presented.