Institutional Repository of Vilnius University
Emocijų atpažinimas pagal elektroencefalogramą naudojantis neuroniniais tinklais /
Abstract
dc:descriptionThis work tests if it is feasible to implement EEG emotion recognition algorithms with a limited amount of channels (electrodes). A Random Forest Regression model and a Convolutional Neural Network are tested. Models are trained on the DEAP affective computing dataset which has been preprocessed: EOG artifacts have been removed, and Fourier Transformations with an overlapping window moved every 0,125 seconds have been applied. 2 main configurations are tested: one with a 10 second overlapping window, the other with a 2 second window. The Random Forest Regressor performs poorly with exact emotional metric classification, however after reducing the amount of possible classes to 2 it reaches an accuracy of 62-65% with a 2 second window and 80-82% with a 10 second window. The convolutional models are able to perform exact classification of emotional metrics. Models with a 2 second window reach accuracies of 76-86%, but can only learn a small subset of the full dataset. The variance in accuracy comes from a differing amount of channels used: adding channels, especially those located on the back of the head, increases model accuracy. The 10 second window models perform with a much higher amount of accuracy: 95- 96.5%. In this case, the amount of channels has no impact, but a more spread out channel group reaches marginally better results.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Sergeda, Paulius,
- Contributors dc:contributor
-
- Valaitis, Vytautas
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD107115663&prefLang=en_US