Literature Review on Phenomenon of Concept Drift and its Handling Approaches

Abstract

Data extracted from dynamic distributions for realworld applications lead to a phenomenon called concept drift. Term concept in “concept drift” primarily refers to an unseen dependency or correlation between input attributes and target variables. There are various factors responsible for the drift in the concept. Such as random noise, change in the distributions of incoming streaming data, variations in the dependency of target variable on input features. This hitch could encounter when the dependency and relationship between input data attributes and target variables would vary gradually. So, the model must be capable of processing this data efficiently and swiftly adapt to these variations. This paper presents an outline of the concept drift research field and provides a review of various kinds of approaches adopted by researchers for mitigating issues related to concept drift.