A comparative study on machine learning approach towards epileptiform eeg signals detection

Oh, Pearly Bei Qing (2017) A comparative study on machine learning approach towards epileptiform eeg signals detection. [Final Year Project Report] (Unpublished)

Preview	PDF A comparative study on machine learning approach...(24 pgs).pdf Download (4MB) | Preview
	PDF (Please get the password by email to repository@unimas.my , or call ext: 082-583914/3973/3933) PEARLY OH BEl QING.pdf Restricted to Registered users only Download (14MB)

Abstract

Electroencephalogram (EEG) signal is extensively used for the diagnosis of various kinds of neurological brain disorders. The classification of normal and abnormal electrical brain spikes through visual inspection is highly subjective and varying across medical experts. Hence, in this project, comparisons between multiple supervised learning approaches are presented, in order to discriminate those epileptiform EEG signals data from non-epileptiform with high generalizability and promising results. Furthermore, both Discrete Wavelet Transform (DWT) and Independent Component Analysis (ICA) are incorporated respectively as a preprocessing stage on reducing dimensionality, besides removing unnecessary noise adequately. Then, a set of statistical extracted features are served as input parameters to various machine learning classifiers, namely Multilayer Perceptron Neural Network (MLPNN), Adaptive Neuro-Fuzzy Inference System (ANFIS) and Support Vector Machine (SVM) respectively with two discrete outputs (normal or epilepsy). As a result, the experimental outputs inferred that the wavelet coefficients which extracted by DWT, have demonstrated as the most well representation of EEG signals. Hence, the ANFIS classifier which trained on these salient features using combination of neural network learning capabilities and fuzzy logic decision making approach, has depicted the highest classification performance accuracy of 99.35% (ICA+ANFIS) and 99.67% (DWT+ANFIS) respectively. Meanwhile, others classifiers namely, MLPNN and SVM, have also proven the diagnostic results with potentially high accuracies, which are 94.39% (ICA+MLPNN), 93.67% (DWT+MLPNN), and 96.22% (ICA+SVM), 94.39% (DWT+SVM) respectively, after tuning the system parameters. Therefore, these findings yield promising outcomes to be presented as a framework for training and testing epileptic prediction on EEG ix data by configuring intelligent devices, so that every patient can be treated in an optimum manner, prior to surgical evaluation.

Actions (For repository members only: login required)

View Item