Modelo de escritura digital basado en señales EEG

Abstract

This document presents the development of software for the detection of P300 evoked potentials with the objective of using them for a speller-type Brain-Computer Interface (BCI). Python, along with the Cyton and Daisy boards from Open BCI, are used for data acquisition, in addition to three other publicly available datasets. The data undergoes pre-processing, which includes independent component analysis, filtering, epoch segmentation, time window division, and feature extraction. For classification, machine learning with simple models is employed, achieving an accuracy of approximately 85% in the best case. This percentage is obtained collectively for all datasets; however, when performing real-time testing, the classification performance decreases significantly.