Semi-supervised support vector machines for data classification with uncertainty

The theory of support vector machines (SVMs) is a new classification technique and has drawn much attention in recent years. The good generalization ability of SVMs is achieved by finding a large margin between two classes. The theory of SVMs has been shown to provide higher performance than traditional learning machines in many applications. However, there are some limitations which restrict its application, for example, it is required every input must be labelled and exactly assigned to one of these two classes without any uncertainty. Sometimes these requirements are too restrictive to be used in practice. In this paper, suppose we are given a set of datasets in which only part of the two-class data is labelled. The labelled input may not exactly belong to any one, but belong to the positive class with a certain probability and/or also belong to the negative class with a certain probability. We propose a new SVMs model to solve the above classification problem. One algorithm has been derived. Some experiments, including an application to monitoring and diagnosing two types of energy loss for utility boilers are presented. The new model extends the application horizon of SVMs greatly