A comparison of PCA, ICA and LDA in EEG signal classification using SVM

Abstract

ÖzetçeEEG işaretlerinin beynin fonksiyonları hakkında çok miktarda bilgi içerdiği bilinmektedir. Epilepsi teşhisinde EEG en önemli bilgi kaynağı olduğu için, birçok araştırmacı EEG işaretlerinden bu amaca uygun bilgi elde etmeye çalışmışlardır. Bu çalışmada sunulan yöntemde, önce EEG işaretlerine öz bağlanımlı (AR) uygulanarak güç spektrumu elde edilmiş, daha sonra elde edilen özellik vektörleri TBA, BBA ve DAA kullanılarak boyut indirgemesi yapılmış; elde edilen değerler destek vektör makinesi (DVM) ile sınıflandırılmaya tabi tutulmuş ve çıkışta epileptik veya değil şeklinde sınıflandırma gerçekleştirilmiştir. Đşaretteki özelliklerin belirlenerek hekime sara tanısında yardımcı olacak, otomatik bir sistem elde edilmesi amaçlanmıştır. DVM ile yapılan EEG sınıflandırmasında DAA'nın daha iyi sonuçlar verdiği ve bu sonuçların hastalık teşhisinde kullanılabileceği görülmüştür.Anahtar Kelimeler: EEG, Epilepsi, AR, TBA; BBA; DAA;Destek vektör makinesi (DVM). AbstractSince EEG is one of the most important sources of information in diagnosis of epilepsy, several researchers tried to address the issue of decision support for such a data. We present a method for classifying epilepsy of full spectrum EEG recordings. In the proposed method, autoregressive (AR) model is used to acquire power spectrum of EEG signals, then dimension of the extracted feature vectors is reduced by using ICA, PCA and LDA, and these vectors used as an input to a support vector machine (SVM) with two discrete outputs: epileptic seizure or not. By identifying features in the signal we want to provide an automatic system that will support a physician in the diagnosing process. It is observed that, SVM classification of EEG signals gives better results and these results can also be used for diagnosis of diseases.

A Brief Review of Brain Signal Monitoring Technologies for BCI Applications: Challenges and Prospects

“…Another common practice for correcting the ocular artifacts (OA) is by using regression analysis [75]. Other methods with Principal Component Analysis (PCA) [76], Independent component analysis (ICA) [77][78][79][80], wavelet based denoising [76][77][78][81][82][83], Wavelet enhanced ICA (wICA) [84][85][86] and wavelet with higher order statistics [87] have also been proposed with varying degrees of success [88]. Jointly using statistical tools like Kurtosis, data improbability, linear trends, spectral pattern with the independent component scalp maps [89] and Kurtosis with Renyi's entropy [90] have also been used to identify artifacts.…”

Patient Specific Seizure Prediction System Using Hilbert Spectrum and Bayesian Networks Classifiers

“…Various feature extraction methods are used, such as spectral power [ 9 ], wavelet coherence [ 19 ], wavelet energy and entropy [ 13 ], short-time Fourier transform [ 20 ], mean phase coherence [ 10 ], and empirical mode decomposition (EMD) [ 21 – 23 ]. Neural networks [ 20 , 24 ] and support vector machines [ 9 , 21 , 25 ] are mostly preferred machine learning algorithms to classify the features extracted from EEG segments. Recently, Bayesian based methods are also employed in seizure prediction systems [ 26 , 27 ].…”

Effect of Two Adjacent Muscles of Flexor and Extensor on Finger Pinch and Hand Grip Force

“…Another type of dimensional reduction technique employed in this study is one of a variant of linear discriminant analysis (LDA) known as uncorrelated linear discriminant analysis (ULDA). LDA as widely discussed in [26], [27], [28], is a linear combination of variables that best separate classes or targets. The idea of proposing ULDA by [29] in 2001, because of the limitation problems in classical LDA requires the scatter matrices to be non singular, and lack of supervision of the dataset decorrelation.…”