Learning Technique Defined Using Concept Drift in Mining System

Abstract: Machine learning approach has got major importance when distribution of data is unknown. Classification of data from the data set arises some problem when distribution of data is unknown. Characterization of raw data relates to whether the data can take on only discrete values or whether the data is continuous. In real world application data drawn from non-stationary distribution, arise the problem of “concept drift” or “non-stationary learning”. Drifting of dataset is often associated with online learning scenario. There are several approaches to track the drift from the dataset; detection of drift has got major research attention. One of the problems of filtering is that it cannot detect concepts change or drift happens as time goes accurately. To deal with the concept drift this paper shows some results of different kind of approach for various kinds of datasets. Detection of drift works for two different levels; warning, and alarm level.

Keywords: Machine Learning, concept drift, Diversity, classification, Bagging, Boosting, Poission Distribution, Ensemble.