A Novel Ensemble Classifier Framework to Preprocess, Learn and Predict Imbalanced Heterogeneous Drifted Data Stream

Organizations nowadays are producing huge size of data stream, whose velocity is extra-ordinary, at a rapid speed. Mining data on stream data is a topic of great interest to the research community and has a great effect of a different types of applications. These applications include banking, communication technologies, field of learning, estimation of forthcoming weather, sensors, social media, and so on. However, Internet of Things (IoT) and smart devices generates enormous volume of data streams that are entirely heterogeneous in nature. The size, rapidity, pattern and category of these data streams are all non-stationary and highly vulnerable to concept drift. It is a significant problem, considering the realm of data mining in stream data. However, one significant and challenging issue in stream data mining is how to deal with the concept drifts, to the same degree, sudden, gradual and incremental drifts, which happens explosively. Additionally, the problem of streaming data classification of nonlinear data stream can only be partially solved by the current methodologies. Most of the works consider fixed size blocks, chunks or windows of homogeneous data stream to address the issues of the concept drift. Thus, a novel ensemble classifier is therefore necessary in order to accurately classify heterogeneous stream of data while swiftly changing to different sorts of concept drifts. This works proposes a novel method, Ensemble Classifier Framework for the detection and learning of concept-drift in addition to the classification in heterogeneous data stream. This work further presents a solution to deal with dynamic occurrence of concept drift.