Identification of novelty and recurrent drift in the streaming environment

In a non-stationary environment, the traditional methods cannot efficiently handle the changing environment to which they are applied. The concept evolution is recently gaining attention by finding changes in the distribution of data (concept drift) where a new class or novelty in a real-time environment. In addition, an identifying concept that previously disappeared and reappeared after a while, known as recurrent drift, is a necessary consideration in a streaming environment. Distinguishing between recurrent drift and concept evaluation becomes even more challenging. Due to increased misclassification errors, several existing works wrongly consider recurrent drift as novelty. To effectively identify the concept drift and concept evaluation and discriminate between recurrent drift and novelty, we develop the NRCD (Novelty, Recurrent, and Concept drift Detection) method. The method first investigates the model coefficients and performs further analysis with features having high coefficient values. With the two windows’ most relevant features, the proposed method builds local and global storage pools to find the changes in the highest coefficients feature’s Interquartile values that are already seen or new features detected in the data stream. For experiment purposes NRCD uses many datasets to compare with various state-of-the-art methods. The proposed work shows an increase in classification accuracy concerning state-of-the-art methods.