Effect of Preprocessing and Sample Size on Support Vector Machine Classification of Nonstationary Signals

M. Brandt-Pearce and D. Reeves (USA)


Support vector machines, classification of nonstationary time series, time-frequency representation.


A support vector machine to be used as a binary classifier for nonstationary time series is analyzed as parameters in cluding the sample size and the type of preprocessing are varied. The algorithm is found to be very sensitive to the data preprocessing and the sample size. The learning ma chine performs better on unprocessed data and short sam ple sizes. The algorithm becomes insensitive to the training sample size if it is sufficiently large.

Important Links:

Go Back