ANÁLISIS DE SEÑALES ELECTROENCEFALOGRÁFICAS PARA CLASIFICAR EMOCIONES UTILIZANDO EL MODELO BIDIMENSIONAL VALENCIA-EXCITACIÓN

LÓPEZ RENTERÍA, JESSICA (2020) ANÁLISIS DE SEÑALES ELECTROENCEFALOGRÁFICAS PARA CLASIFICAR EMOCIONES UTILIZANDO EL MODELO BIDIMENSIONAL VALENCIA-EXCITACIÓN. Maestría thesis, UNIVERSIDAD AUTONOMA DE CHIHUAHUA.

[img]
Vista Previa
Text
Tesis Final (Ing. Jessica López Rentería 259719).pdf

Download (4Mb) | Vista Previa

Resumen

Emotions are defined as complex mental states, generated from the reaction of human being to internal and external stimuli that produce changes at a physiological level. Which can be monitored for the purpose of identifying the different emotional human states. Currently there are few public domain databases that can be used to analyze emotions in electroencephalographic signals (EEG), which is why in the current research work a database was created using the AURA software and electroencephalograph. This with the support of the Mirai Innovation Research Institute Laboratory in Osaka, Japan. The database contains information from 14 participants from different parts of México, who were shown a series of audiovisual marketing stimuli focused on the Generation of 4 different emotions: happiness, sadness, calm and anger. This work presents graphs, tables and results that allow to evaluate and analyze the performance of computational algorithms for the classification of emotions through the extraction and selection of Features in EEG Signals in two databases: the database created and the DEAP public domain database. The features extraction was based on time and time-spectrum domain, using the raw EEG signal and the Wavelet decomposition respectively. The Random Forest algorithm achieved the best performance in terms of the classification problem in the created database with 99% average accuracy in the 4-emotion classification. However, the results obtained from the Support Vector Machine and Random Forest algorithm for the DEAP database were lower (38% and 34% respectively) compared to the state of the art. Therefore, an extra analysis was developed based on a binary classification which was performed through the algorithm of principal component analysis (PCA) and the algorithm described above. The result in the dimensions of valence and arousal was 67.84% and 68.42% respectively. In addition, a Multilayer Perceptron (MLP) was implemented to establish a comparison of the obtained results. The average accuracy percentage of MLP was 56.04% for the valence dimension and 58.11% for the arousal dimension. Emotions analysis is an area that requires more scientific research, for the future development of real-time applications.

Tipo de Documento: Tesis (Maestría)
Palabras Clave: EEG, machine learning, DEAP, principal component analysis, marketing.
Clasificación temática: Q Science > QA Mathematics
Usuario Remitente: Admin Administrador del Respositorio
Depositado: 07 May 2021 16:22
Ultima Modificación: 07 May 2021 16:22
URI: http://repositorio.uach.mx/id/eprint/324

Actions (login required)

Ver Objeto Ver Objeto

Universidad Autonoma de Chihuahua