A novel data reduction method: Distance based data reduction and its application to classification of epileptiform EEG signals

摘要: Abstract Objective Data reduction methods are a crucial step affecting both performance and computation time of classification systems in pattern recognition applications such as medical decision making systems, intelligent control, data clustering. The aim this study is to increase the accuracy decrease classifier system on epileptiform EEG signals. Methods In study, we have proposed novel method based distances between groups double all dataset applied feature extraction including autoregressive (AR), discrete Fourier transform (DFT), wavelet (DWT), distance reduction, C4.5 tree been combined classify As part AR, DFT, DWT used determine features about signals epileptic seizure patients eyes open volunteers. pre-processing part, that firstly by us has reduce determined spectral analysis (AR, DWT). final called classification, reduced Results To validate test accuracy, sensitivity, specifity analysis, time, 10-fold cross-validation, 95% confidence intervals study. Six different signal. These (i) combining DFT (DCT), (ii) DCT, (iii) AR (iv) (v) (vi) DWT, DCT. accuracies times obtained these 99.02% – 79 s, 99.12% 47 s, 99.32% 65 s, 98.94% 45 s, 92.00% 52.06 s, 89.50% 29.9 s. Conclusions results shown produced very promising with respect for classifying Also, hybrid can be detect seizure.