Selection and optimization of peak features for event-related eeg signals classification / Asrul bin Adam

Asrul, Adam (2017) Selection and optimization of peak features for event-related eeg signals classification / Asrul bin Adam. PhD thesis, University of Malaya.

[img] PDF (The Candidate's Agreement)
Restricted to Repository staff only

Download (1593Kb)
    PDF (Thesis (Ph.D.)
    Download (3715Kb) | Preview


      The classification of desired peaks in event-related electroencephalogram (EEG) signals becomes a challenging problem for brain signals researchers. The reasons are mainly because of the peak in the signals have been contaminated with various noises, the nature of non-stationary EEG signals, many peaks candidates in the signals, and the peak features relative to the baseline amplitude, time, and different users. Many peak classification algorithms have been introduced for various EEG signals applications. However, the developed algorithms only consider the selected features from a peak model based on the understanding of the EEG signals characteristics. The utilization of different existing models cannot assure giving the best classification performance for other event-related EEG signals applications. For a fair performance evaluation, the selection of the best peak model requires experimental exploration by using a common and unbiased classification approach. This thesis aims to provide a high and good generalized peak classification performance through the application of an optimization approach with the advantageous of a common classification method for finding a new optimal combination of peak features. At first, a peak classification algorithm is developed based on the general following processes including peak candidate identification, feature extraction, and classification. Four different existing peak models with the associated features and full features set model are considered as inputs to the classifier. The four peak models are named as Dumpala, Acir, Liu, and Dingle models whereas the full features set model consists of 16 peak features. Three event-related EEG signals that recorded from 30 voluntary of healthy subjects, namely as a single eye blink, double eye blink, and eye movement signals are employed. All subjects are instructed to direct their eye blinks and horizontal gaze in response to a voice cue. In the preliminary study, the algorithm is evaluated on the four different peak models of the three EEG signals using the artificial neural network (ANN) with particle swarm optimization (PSO) as learning algorithm. Unfortunately, the ANN classification method cannot provides the fast learning speed once it integrates with the PSO. The study continued with other classification technique which is neural network with random weights (NNRW). Next, four recently introduced optimization algorithms are employed as feature selector, namely as 1) angle modulated simulated Kalman filter (AMSKF), 2) binary simulated Kalman filter (BSKF), 3) local optimum distance evaluated simulated Kalman filter (LocalDESKF), and 4) global optimum distance evaluated simulated Kalman filter (GlobalDESKF). This study resulted in a new generalized model based on the best performance among the four novels simulated Kalman filter (SKF) approaches. The new generalized models with the associated features that are selected using the novel feature selection approaches have substantially improved the performance of the existing models. The proposed models and NNRW method in this thesis perform at par with the existing related studies of epileptic EEG events classification.

      Item Type: Thesis (PhD)
      Additional Information: Thesis (Ph.D.) -– Faculty of Engineering, University of Malaya, 2017
      Uncontrolled Keywords: Electroencephalogram (EEG) signal; Algorithms; Neural network with random weights (NNRW); Signals classification
      Subjects: T Technology > TA Engineering (General). Civil engineering (General)
      T Technology > TK Electrical engineering. Electronics Nuclear engineering
      Divisions: Faculty of Engineering
      Depositing User: Mr Mohd Safri Tahir
      Date Deposited: 08 Jul 2017 09:53
      Last Modified: 08 Jul 2017 09:53

      Actions (For repository staff only : Login required)

      View Item