A Multi-partition Multi-chunk Ensemble Technique to Classify Concept-Drifting Data Streams

We propose a multi-partition, multi-chunk ensemble classifier based data mining technique to classify concept-drifting data streams. Existing ensemble techniques in classifying concept-drifting data streams follow a single-partition, single-chunk approach, in which a single data chunk is used to train one classifier. In our approach, we train a collection of v classifiers from r consecutive data chunks using v -fold partitioning of the data, and build an ensemble of such classifiers. By introducing this multi-partition, multi-chunk ensemble technique, we significantly reduce classification error compared to the single-partition, single-chunk ensemble approaches. We have theoretically justified the usefulness of our algorithm, and empirically proved its effectiveness over other state-of-the-art stream classification techniques on synthetic data and real botnet traffic.