Joint Split Criterion Based Data Stream Classification Technique

Anushree R Bhat Kanak, Prof. H R Shashidhara

The quantity of data that needs to be analyzed has lead to a new field data stream mining. The goal of most of data stream mining application is to predict the class or value of new instances in data stream which gives some knowledge about the class members. Data classification is one of the important techniques used in data mining.This paper involves developing a new method for constructing decision tree stream data. Along with the existing methods, misclassification error is used as another measure to build hybrid technique which provides an accurate measure of impurity. A hybrid technique is developed using Chi-square and misclassification error to calculate impurity measure. Impurity measure is then used in split criteria to build the decision tree classifier to classify the stream data. Accuracy is calculated during the data streaming process. It is shown that the proposed system i.e. the hybrid technique provides satisfactory accuracies at any time of data stream processing.