An optimized deep learning approach based on autoencoder network for P300 detection in brain computer interface systems

Background. Brain computer interface (BCI) systems by extracting knowledge from brain signals provide a connection channel to the outside world for disabled people, without physiological interfaces. Event-related potentials (ERPs) are specific type of electroencephalography and P300 is one most important ERP components. The critical part P300-based BCI classification step. In this research, an approach proposed based on novel machine learning methods using convolutional neural networks (CNN) autoencoder networks. Methods. pre-processing step, selection, data augmentation (by ADASYN method), filtering base-line drift were done. Then, in four different CNN classifiers including CNN1D, CNN2D, CNN1D_Autoencoder, CNN2D-Autoencoder used classification. Results. After implementation tuning networks, 92% as best accuracy was achieved CNN2D_Autoencoder. This result with considerable tradeoff between complexity stability. Conclusion. acquired results emphasize ability deep approve advantage them systems. Furthermore, versions more stable have faster convergence. Meanwhile, suitable method even ERPs sustaining premier feature space copying data. Practical Implications. Our can increase detection simultaneously reduce volume model. Consequently, they improve character recognition P300-speller generally amyotrophic lateral sclerosis (ALS) patients.