An Approach for Concept Drifting Streams: Early Dynamic Weighted Majority

摘要

Techniques for handling drifting concepts are very significant for many real time applications. In this paper, a novel methodology has proposed, Early Dynamic Weighted Majority, based on assigning weight to the classifiers in the ensemble. It dynamically updates the weight by increasing the weight when the local prediction of an expert is correct which is in contrast with a decreasing weight feature as in the earlier approaches. The system is a collection of experts that make a prediction based on weighted-majority voting. Empirical results prove that the proposed system is the best in managing drifting concepts in the underlying conceptual distribution.