Prior knowledge in supervised classification models

Classifiers built through supervised learning techniques are widely used in experimental sciences. Examples are neural networks, decision trees and support vector machines. Recently, when knowledge is formalized as a set of linear constraints, an extension of those classifiers has been proposed. The resulting classifiers have lower complexity and half the misclassification error, with respect to the original methods. In this work, we show how to extract knowledge from data to enhance classification models. The overall methods guarantee that the number of points in the training set is not increased and the resulting model does not over-fit the problem. A case study is provided, based on a cancer data set taken from the literature.