Título: Stacking Ensemble Learning Approaches Applied to Emotional State Classification
Autores: Luiza Scapinello Aquino and Leandro dos Santos Coelho.
Resumo:
The ability to automatically recognize human emotions is a wide field of research that supports both psychology and engineering, through the improvement of human-computer interface systems, and psychiatry, assisting in diagnosis, and treatment decision making of mental diseases. To deal with big amounts of information, artificial intelligence (AI) approaches are often adopted, such as machine learning (ML), which enables computers to learn and adapt to new situations automatically. This paper aims to compare some ML approaches in classifying the human emotional states from electroencephalogram examination signals through software implementation. The dataset used in this study, known as DEAP (Database for Emotion Analysis using Physiological Signals), comes from a set of exams available for the public by a researcher group from four European Universities. A comparison between the methods decision tree, support vector machine (SVM), and convolutional neural networks (CNN) was made in terms of performance metrics. Moreover, unlike other works focused on these concepts, combinations of these architectures were used in this study to find an accurate result, through the stacking ensemble learning approaches applying in human emotion classifications. Therefore, eight classification methods were applied, three of which have no history of use in this application previously. K-fold and cross validation were used in order to estimate the algorithms capacity of generalization. The best results were obtained with the ensemble of three base methods, with an accuracy of 0.946, and the combination of CNN with SVM obtained 0.922 of accuracy metric.
Palavras-chave:
machine learning, stacking, ensemble learning, emotion classification, electroencephalography.
Páginas: 8
Código DOI: 10.21528/CBIC2021-15
Artigo em pdf: CBIC_2021_paper_15.pdf
Arquivo BibTeX: CBIC_2021_15.bib