SVM based Feature Extraction for Novel Class Detection from Streaming Data

Abstract

World have huge amount of data. Data stream classification contain several problem such as Infinite Length , Concept Drift ,Concept Evolution and Feature Evolution. Infinite Length means data available in huge amount and it is difficult to store all historical data for training. Concept Evolution occurs as a result of new classes evolving in stream. Concept Drift occurs as a result of changes in underlying concepts. Feature Evolution occurs as new feature arises. Traditional data stream classifier only addresses Infinite Length and Concept Drift. In this paper we propose ensemble classification framework where each classifier is equipped with novel class detector to address Concept Drift and Concept Evolution. Also increases accuracy of novel class detection techniques by using SVM based polynomial kernel.