Classificação de emoções em sinais de EEG utilizando técnicas de aprendizado de máquina

• Ano de defesa: 2017
• Autor(a) principal: Simão, Vancley Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Programa de Pós-Graduação em Engenharia de Sistemas e Automação; UFLA; brasil; Departamento de Engenharia
• Palavras-chave em Português: Emoções – Classificação – Algorítmos computacionais; Aprendizado de máquina – Modelos; Redes neurais (Computação); Florestas aleatórias (Computação); Máquinas de vetores de suporte; Emotions – Classification – Computer algorithms; Machine learning – Models; Neural networks (Computer science); Random forest (Computer science); Support vector machines; Análise de Algoritmos e Complexidade de Computação; Sistemas de Computação
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/28230

Emotions play a fundamental role in human daily lives, having influences over decisions and even communications. Understanding how emotions are characterized and how they can be identified are of utmost importance to understand how humans behave. Various methods have been proposed for emotion classification, however researches that correlates discrete emotions and brain activity patterns are still being investigated. Thus this master’s dissertation presents a methodology for discrete emotion classification in EEG signals using machine learning algorithms.During the development of this work various models have been created using machine learning algorithms where accuracy metrics have been noted to be compared and thus select the best classification model. For the development of the models, the DEAP dataset was used. Once the dataset was unbalanced, the balancing algorithms SMOTE and ADASYN were evaluated in their capacity to balance data and contribute to the improvement of the classification models.The main contributions of this dissertation are: the development of the discrete emotion classification models in EEG signals and the evaluation of the machine learning and the data balancing algorithms.After developed the models, the results showed that the Random Forest model with the ADASYN algorithm for data balancing had the best results with an average accuracy of 89.22%. The models built with the SVM algorithm and the ADASYN technique for balancing the data also showed good results with an average accuracy of 87.36%. Finally, the models with Neural Networks and SMOTE algorithms showed the worst results with an average accuracy of 68.56%.Therefore, the results showed that for the discrete emotion classification in EEG signals, the models using the ADASYN algorithm for data balancing with the Random Forest algorithm for classification were superior when compared to the other tested models.