An approach to handle novel classes: Weighted Novel Class Detection algorithm

Many algorithms have been designed in the past that tackle the problem of concept drift in data streams. We present a new approach Weighted Novel Class Detection (WNCD), a diversified ensemble approach that combines the concepts of ensemble learning, instance weighting and diversity, for handling drifting concepts and detection of novel class instances. Based on the empirical results using various artificial and real time datasets, WNCD gives better performance as compared to the existing online approaches in handling concept drift and novelty detection.