Estimation of Emotions and Mental Concentration using Deep Learning Techniques

Abstract

The purpose of this work is to evaluate the brain waves of humans with deep learn- ing methods and evolutionary computation techniques, and to verify the performance of applied techniques. In this thesis, we apply well–known metaheuristics and Artificial Neural Networks for classifying human mental activities using electroencephalographic signals. We developed a Brain–Computer Interface system that is able to process elec- troencephalographic signals and classify mental concentration versus relaxation. The system is able to automatically extract and learn representation of the given data. Based on scientific protocols we designed the Brain–Computer Interface experiments and we created an original and relevant data for the industrial and academic community. Our experimental data is available to the scientific community. In the experiments we used an electroencephalographic based device for collecting brain information form the subjects during specific activities. The collected data represents brain waves of subjects who was stimulated by writing tasks. Furthermore, we selected the best combination of the input features (brain waves information) using the following two metaheuristic techniques: Simulated Annealing and Geometric Particle Swarm Optimization. We applied a specific type of Artificial Neural Network, named Echo State Network, for solving the mapping between brain information and subject activities. The results indicate that it is possible to estimate the human con- centration using few electroencephalographic signals. In addition, the proposed system is developed with a fast and robust learning technique that can be easily adapted accord- ing to each subject. Moreover, this approach does not require powerful computational resources. As a consequence, the proposed system can be used in environments which are computationally limited and/or where the computational time is an important issue.