pdi-Bagging: A Proposal of Possibilistic Data Interpolation Bagging

We propose a new possibilistic data interpolation bagging (pdi-Bagging), which improve the discriminant rate of checking data by adding virtually generated data to the training data. In this paper, we propose a new method to generate virtual data not only for misclassified data but also around correct classified data, and a new method for determining the output class of virtual data. In addition, we discuss the usefulness of pdi-Bagging using Student's t-test and Tukey's HSD test in numerical examples, and finally, we also discuss its application to a vehicle type discrimination system in the large-scale outdoor parking lot.