Electroencephalography (EEG) is one of the most important signals recorded from
humans. It can assist scientists and experts to understand the most complex part of the
human body, the brain. Thus, analysing EEG signals is the most preponderant process
to the problem of extracting significant information from brain dynamics. It plays a
prominent role in brain studies. The EEG data are very important for diagnosing a
variety of brain disorders, such as epilepsy, sleep problems, and also assisting
disability patients to interact with their environment through brain computer interface
(BCI). However, the EEG signals contain a huge amount of information about the
brain’s activities. But the analysis and classification of these kinds of signals is still
restricted. In addition, the manual examination of these signals for diagnosing related
diseases is time consuming and sometimes does not work accurately. Several studies
have attempted to develop different analysis and classification techniques to categorise
the EEG recordings. The analysis of EEG recordings can lead to a better understanding of the cognitive
process. It is used to extract the important features and reduce the dimensions of EEG
data. In the classification process, machine learning algorithms are used to detect the
particular class of EEG signal based on its extracted features. The performance of these
algorithms, in which the class membership of the input signal is determined, can then
be used to infer what event in the real-world process occurred to produce the input
signal. The classification procedure has the potential to assist experts to diagnose the
related brain disorders. To evaluate and diagnose neurological disorders properly, it is
necessary to develop new automatic classification techniques. These techniques will
help to classify different EEG signals and determine whether a person is in a good
health or not. This project aims to develop new techniques to enhance the analysis and
classification of different categories of EEG data. A simple random sampling (SRS) and sequential feature selection (SFS) method
was developed and named the SRS_SFS method. In this method, firstly, a SRS
technique was used to extract statistical features from the original EEG data in time
domain. The extracted features were used as the input to a SFS algorithm for key features selection. A least square support vector machine (LS_SVM) method was then
applied for EEG signals classification to evaluate the performance of the proposed
approach. Secondly, a novel approach that combines optimum allocation (OA) and spectral
density estimation methods was proposed to analyse EEG signals and classify an
epileptic seizure. In this study, the OA technique was introduced in two levels to
determine representative sample points from the EEG recordings. To reduce the
dimensions of sample points and extract representative features from each OA sample
segment, two power spectral density estimation methods, periodogram and
autoregressive, were used. At the end, three popular machine learning methods
(support vector machine (SVM), quadratic discriminant analysis, and k-nearest
neighbor (k-NN)) were employed to evaluate the performance of the suggested
algorithm. Additionally, a Tunable Q-factor wavelet transform (TQWT) based algorithm was
developed for epileptic EEG feature extraction. The extracted features were forwarded
to the bagging tree, k-NN, and SVM as classifiers to evaluate the performance of the
proposed feature extraction technique. The proposed TQWT method was tested on two
different EEG databases. Finally, a new classification system was presented for epileptic seizures detection in
EEGs blending frequency domain with information gain (InfoGain) technique. Fast
Fourier transform (FFT) or discrete wavelet transform (DWT) were applied
individually to analyse EEG recording signals into frequency bands for feature
extraction. To select the most important feature, the infoGain technique was employed.
A LS_SVM classifier was used to evaluate the performance of this system. The research indicates that the proposed techniques are very practical and effective
for classifying epileptic EEG disorders and can assist to present the most important
clinical information about patients with brain disorders. |