Abstract
This paper abstracts the contents of the PhD dissertation which has been recently defended by the author. Machine learning is the area of computer science that is concerned with the question of how to construct computer programs that automatically improve with experience. Recently, there have been important advances in the theoretical foundations of this field. At the same time, many successful applications have been developed: systems for extracting information from databases (data mining), applications to support decisions in medicine, telephone fraud and network intrusion detection, prediction of natural disasters, email filtering, document classification, and many others. This thesis introduces novel supervised learning methods that produce accurate and comprehensible models from past experiences which minimise the costs of generation and the costs of application.