A distance-based classifier using dissimilarity based on class conditional probability and within-class variation

According to the rapid increase of data, the needs of intelligent data analysis and classification are also increasing. Though there have been developed various methods of classifying given data set into several pre-defined patterns, the distance-based classifier such as nearest neighbor classifier is still one of the most popular methods due to its simplicity and adaptability. However, in order to obtain good performances in practical applications, it is important to choose an appropriate distance measure considering the purpose of task and the distributional properties of data set. In this paper, we propose a new measure of similarity based on two probability densities: The classconditional probability and the probability of within-class variation. Through statistical estimation of the probability densities using training set, it is possible to obtain an optimized measure for the given data. The efficiency of the proposed measure is confirmed by computational experiments on a few pattern recognition problems using benchmark data sets.