Classification algorithm of streaming signals based on the online support vector machine

The work proposed a modification of support vector machines (SVM) to train and classify in real time (online) streams of data. The algorithm is tested on the data handwriting figures and shown that its error is comparable to SVM direct solution error. Speed and support vectors number of proposed SVM algorithm is smaller than in other known SVM implementations. Finally, a ternary classificator for 2-class problem is proposed which shows better results than binary.