Learning Framework for Non-stationary and Imbalanced Data Stream

Abstract—Although learning on non-stationary data and imbalanced data have been extensively studied in the literature separately, however little work has been done to tackle the imbalanced issue on nonstationary data stream as the joint probability distribution between the data and classes changes with time and may results skewed class distribution. Especially in airlines delay detection, data sources are dynamic generated at high speed in real time, type of delay activity changes with time and in each chunk of stream, delay detection instances are less so concept drift and class imbalanced issues arises simultaneously. Through this research work we propose an ensemble based incremental learning approach towards non-stationary imbalanced data stream. Keyword-Concept Drift, Ensemble, Imbalanced Data, Incremental Learning,Non-stationary Data