Abstract
In recent years, a number of classification algorithms using fuzzy decision trees have been proposed. A key issue encountered in all of these algorithms is how to select the division feature. In traditional algorithms, it is common to use the information gain criterion. There is a claim that the use of information-based criteria leads to an unfair selection of features that have large number of values while not being suitable for division. This problem causes to decrease accuracy of classification in stream data. However, other algorithms for classifying stream data and extracting fuzzy rules have been proposed but the problem of unfairly selecting features with large amount of values still persists. This paper, present a new algorithm to solve problems of selecting splitting feature in fuzzy decision tree for classification of stream data. In proposed algorithm, we extend multi flexible fuzzy decision tree (MFlexDT) with chi-square based fuzzy partitioning of values. In this paper, we also extend traditional chi-square statistical to fuzzy chi-square statistical data distribution. In evaluating the proposed algorithm, tree depth and accuracy are two factors that affect performance. Experimental results show that the new algorithm has been able to improve the performance of other algorithms.
•Unfair selection of attributes with many number of values, causes to increase depth of tree.•The MFlexDT algorithm has been extended to solve the problem mentioned above.•Experimental results show that in addition to improving performance of MFlexDT, in decreasing the depth of tree the new algorithm was also able to increase kappa static and accuracy.