Ensemble framework for concept drift detection and class imbalance in data streams

Many data mining application generate data in the form of streams called as streaming data and they arrive continuously. The distribution of data changes over time for the streaming data. The online ensemble learning method is used to handle the change in the underlying distribution of data called as concept drift, which gives the timely response for incoming data instances. Although many methods have been proposed for concept drift detection, data streams pose challenge in learning from concept drift with class imbalance, which exist in real world application such as intrusion detection and fault detection. Therefore, it is a significant challenge for the machine learning community to learn from the drifting and imbalanced data stream. In this paper, Ensemble Based Enhanced Early Drift Detection Model and Random Resampling Technique for online learning is proposed which detects the drift based on average error rate and standard deviation. The class imbalance is handled in the data stream by generating the synthetic data using the random resampling technique and the concept drift adaptation is done using ensemble classifiers. The proposed ensemble method can handle both concept drift and class imbalance with 98.52% accuracy.