A novel and incremental classification algorithm
Hüseyin Özkan, Ozgun S. Pelvan, Arda Akman, Süleyman S. Kozat · 2012
In this paper, using “context tree weighting method”, a novel classification algorithm is proposed for real time machine learning applications, which is mathematically shown to be “competitive” with respect to a certain class of algorithms. The computational complexity of our algorithm is independent with the amount of data to be processed and linearly controllable. The proposed algorithm, hence, is highly scalable. In our experiments, our algorithm is observed to provide a comparable classification performance to the Support Vector Machines with Gaussian kernel with 40~1000× computational efficiency in the training phase and 5~35× in the test phase.