Izlučivanje značajki iz elektroencefalograma i klasifikacija u svrhu detekcije pospanosti primjenom umjetnih neuronskih mreža

master's thesis

• Title (english): Feature Extraction from Electroencephalogram and Drowsiness Classification Using Artificial Neural Networks
• Author: Petra Franjić
• Mentor: Mario Cifrek (mentor)
• Committee member: Mario Cifrek (predsjednik povjerenstva)
• Committee member: Alan Jović (član povjerenstva)
• Committee member: Željka Lučev Vasić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2017-07-04, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu ispitana je uspješnost razlicitih tipova neuronskih mreža u
zadatku detekcije pospanosti na temelju jednog kanala elektroencefalografskog (EEG)
zapisa. U sklopu rada, na temelju EEG signala ekstrahiran je skup znacajki u razlicitim
domenama analize. Postupkom selekcije znacajki generirani su razliciti podskupovi
znacajki koji su zatim ispitani u radu sa skupom klasifikatora. Kao najbolji klasifikator
istaknula se mreža s povratnim slojevima koja je u kombinaciji s 4 ekstrahirane znacajke
postigla srednju ukupnu tocnost klasifikacije razlicitih stanja svijesti u iznosu 83,3% i
Cohenov kappa koeficijent 0,73. Isti klasifikator je u kombinaciji s 20 znacajki ostvario i
najbolju postignutu detekciju pospanosti, koja je iznosila 55%. Ispitivanje je provedeno
na javno dostupnoj Physionet Sleep-EDF [Expanded] bazi polisomnografskih snimaka.

Abstract (english)

The aim of this thesis was to evaluate the performance of different
types of neural networks with regard to the task of drowsiness detection based on a
single electroencephalographic (EEG) channel. Firstly, a set of features was formed
based on feature extraction in distinct domains of analysis. Then, different subsets of
features were generated through a feature selection procedure. Finally, the performance
of classifiers of interest was examined when they were used in combination with the
generated feature subsets. The results singled out the layer-recurrent neural network as
the best-performing classifier, achieving 83,3% overall classification accuracy of different
states of wakefulness with Cohen’s kappa coefficient 0,73. This classifier also achieved
the best drowsiness detection performance, reaching 55% recall when it was used with
20 features. Data were obtained from the publicly available Physionet Sleep-EDF
[Expanded] database of polysomnographic recordings.

• Keywords: elektroencefalografija; detekcija pospanosti; neuronske mreže; povratne neuronske mreže; obrada biomedicinskih signala
• Keywords (english): electroencephalography; drowsiness detection; neural networks; recurrent neural networks; biomedical signal processing
• Language: croatian
• URN:NBN: urn:nbn:hr:168:906597
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-13 20:17:37