Detection of Novel Class for Data Streams

Data stream mining is a process of extracting the information from continuously coming rapid data records. Data stream can be viewed as an ordered sequence of instances appears at time varying. Data stream classification has three major problems: infinite length, concept drift and concept evolution or arrival of novel class. In this paper, we propose a new approach for detection of novel class using decision tree classifier that determine whether a new or unseen data instance belongs to an existing class or novel class. It builds decision tree from training datasets, so the tree represents the most recent concept by constantly updating it. The experiment on different datasets from UCI machine learning repository evaluate the efficiency of proposed approach for detecting novel class under dynamic attribute set and classification accuracy by comparing with traditional data mining classifiers.