A Review of Classification and Novel Class Detection Technique of Data Streams

Abstract

Stream data classification suffered from a problem of infinite length, concept evaluation, feature evaluation and data drift. Data stream labeling is more challenging than label static data because of several unique properties of data streams. Data streams are suppose to have infinite length, which makes it difficult to store and use all the historical data for training. Earlier multi-pass machine learning technique is not directly applied to data streams. Data streams discover concept-drift, which occurs when the discontinue concept of the data changes over time. In order to address concept drift, a classification model must endlessly adapt itself to the most recent concept. Various authors reduce these problem using machine learning approach and feature optimization technique. In this paper we present various method for reducing such problem occurred in stream data classification. Here we also discuss a machine learning technique for feature evaluation process for generation of novel class.