Feature Selection Algorithms to Classify Multichannel Uterine Magnetomyography Signals

T. Ananda Babu and Dr.P. Rajesh Kumar

The detection of labor is crucial for proper care to the infants and mothers as most of the complications happened at the onset of labor. The uterine activity measured by using magnetomyographic signals analyzed in this research for the detection of term labor. The MMG signals from Physionet mmgdb database decomposed with discrete wavelet transform. The decomposition performed up to six levels to compute the features, i.e. energy, waveform length, standard deviation, entropy and variance from the transform coefficients. Significant features were selected using particle swarm optimization algorithm and fed to four different classifiers for the labor assessment. The performance of each classifier calculated by using different mother wavelets. The support vector machine classifier trained with PSO selected features is good at classifying the pregnancy and labor with an accuracy of 96.1116%. The results obtained from the research will be helpful in term labor assessment and understanding the parturition process.

Volume 11 | Issue 8

Pages: 211-219