Detection & Management of Concept Drift

The ability to correctly detect the location and derive the contextual information where a concept begins to drift is essential in the study of domains with changing context. This paper proposes a top-down learning method with the incorporation of a learning accuracy mechanism to efficiently detect and manage context changes within a large dataset. With the utilisation of simple search operators to perform convergent search and JBNC with a graphical viewer to derive context information, the identified hidden context are shown with the location of the disjoint points, the contextual attributes that contribute to the concept drift, the graphical output of the true relationships between these attributes and the Boolean characterisation which is the context