A Consolidated Study On Advanced Classification Techniques Used On Stream Data

With the era of IOT, every device is bound to generate data and every digital footprint is noted. This advances in the technology gave rise to data generation at large stature and rapid speed. Such data are termed as Data Streams. They are extremely dynamic and always changing, which leaves them open to concept drift, or abrupt changes in the underneath data distribution. To maintain the accuracy and reliability of mining models over time, it is crucial to detect concept drift in real-time. Classification-based methods employ a classifier model to forecast the class label of incoming data events and use accuracy monitoring to detect drift points. Some of the commonly used classification-based methods for concept drift detection include ADWIN, DDM, EDDM, Hoeffding Trees, and Leveraging Bagging. The method chosen is defined by the specific requirements of the application as well as the characteristics of the data stream. This paper gives an overview of various approaches to aid scholars and practitioners in understanding the theories and methods related to concept drift detection in data stream mining utilizing classification-based approaches.