A Modified Decision Tree Algorithm for Uncertain Data Classification

Abstract

The classifications of uncertain data become one of the tedious processes in the data mining domain. The uncertain data are contains tuples with different probability distribution and thus to find similar class of tuples is a complex process. When we consider uncertain data, the feature vector will not be a single valued but a function. Recently, different methods are proposed on decision tree based uncertain data classification with binary based operation on the decision tree. When multiclass data are given to the decision tree, their algorithm has to give repeated calculation to produce the probability distribution matching the class labels, thus time and memory utilization will be high for the particular algorithm. In this paper, we have intended to propose a classification method for uncertain data based on the decision tree. The proposed approach concentrates on an adaptive averaging method, where we have incorporated mean and median of the tuple to produce the feature value that will be used in the decision tree for decision making. Then a probability calculation is executed to find the relevance of tuple with respect to a class. If the calculated probability value is similar to a particular probability distribution, then the tuple is marked to that particular class. Thus, we produce a decision tree with c number of leaf nodes, where c is the number of class labels in the training data. The test data is subjected to the trained decision tree to obtain the classified data. . The experimental analysis are conducted for evaluating the performance of the proposed approach. The vehicle dataset and segment dataset from the UCI data repository is selected for the performance analysis. The results from the experimental analysis showed that the adaptive method has achieved a maximum average accuracy of 0.997 while the existing approach achieved only 0.985