Adaptive Support Vector Machine for Time-Varying Data Streams Using Martingale

A martingale framework is proposed to enable support vector machine (SVM) to adapt to time-varying data streams. The adaptive SVM is a one-pass incremental algorithm that (i) does not require a sliding window on the data stream, (ii) does not require monitoring the performance of the classifier as data points are streaming, and (iii) works well for high dimensional, multi-class data streams. Our experiments show that the novel adaptive SVM is effective at handling time-varying data streams simulated using both a synthetic dataset and a multi-class real dataset.