Decision rule for pattern classification by integrating interval feature values

Pattern classification based on Bayesian statistical decision theory needs a complete knowledge of the probability laws to perform the classification. In the actual pattern classification, however, it is generally impossible to get the complete knowledge as constant feature values are influenced by noise. Therefore, it is necessary to construct more flexible and robust theory for pattern classification. In this paper, a pattern classification theory using feature values defined on closed interval is formalized in the framework of Dempster-Shafer measure. Then, in order to make up the lack of information, an integration algorithm is proposed, which integrates the information observed by several information sources with considering source values