A Study on a Monotonic ε Support Vector Machine Model

Abstract

Pattern recognition is a very active field of research intimately bound to machine learning Classification is a part of this area that aims to build classifiers that can determine the class of an input pattern Data mining techniques have been applied to solve classification problems in real world applications Support vector machines (SVMs) have recently been introduced for classification and quickly became the state-of-the-art Its excellent ability is the focus of research in machine learning There are human experts in many fields whose knowledge can significantly influence the effectiveness classification Now the incorporation of prior knowledge into SVMs has become the key element in improving the performance of SVM’s in many applications In recent studies researchers attempted to improve the performance of SVMs by using the distance metric learning algorithm and the result was the εSVM The εSVM optimizes the radius-margin ratio error and is thus simpler than traditional SVMs because it does not involve feature selection weighting and multiple kernel learning In this study we explore the incorporation of prior knowledge in the form of monotonicity constraints in an εSVM Our classification model is implemented by constructing monotonicity constraints into anεSVM and determining the contribution of different information Experiment results show that the proposed model which considers the monotonicity of prior knowledge and contribution of different data performs better than the original εSVM model in solving classification problems

Date of Award	2014 Jul 7
Original language	English
Supervisor	Sheng-Tun Li (Supervisor)