Configuração Automática de Hiperparâmetros em Sistemas BCI-MI de Sub-bandas

This work applies Bayesian optimization in the automatic adjustment of hyperparameters for classification of motor imagery in brain-computer interfaces. The strategy consists of modeling the accuracy of the classifier according to the hyperparameters and seeking an optimal personalized setup. Simulations from public data resulted in an average performance gain of up to 7.9% compared to fixed configuration models, indicating that auto setup can minimize the EEG signal variability impact on performance and contribute to a more generalist system.