Using HDDT to avoid instance propagation in unbalanced and evolving data streams

Hellinger distance has been successfully used as a tree splitting criterion in Hellinger Distance Decision Trees [10] (HDDT) for unbalanced static datasets. In unbalanced data streams, state-of-the-art techniques use instance propagation and standard decision trees to cope with the unbalanced problem. However it is not always possible to revisit/store old instances of a stream. We solve this problem using HDDT in data streams. In this paper we show how HDDT can be successfully applied in unbalanced and evolving stream data, leading to improved predictive accuracy and speed. We then use a Hellinger weighted ensemble of HDDTs to combat concept drift and increase accuracy. We test our framework on several streaming datasets with unbalanced classes and concept drift. HDDT, Hellinger distance, Unbalanced data, Streaming data, Concept drift, Fraud detection.