Adaptive Chirplet Transform-Based Machine Learning for P300 Brainwave Classification

Abstract

Within brain-computer interface (BCI) research, classification of event related potentials (ERP s) is of great interest as a method of understanding the brain, as well as for human-computer communication. In particular, investigation of P300 brainwaves is relevant due to its ease of use in the context of BCI. Though high-accuracy models have been developed for individual subjects under laboratory controlled conditions, further work is required to improve the robustness of the models especially for more varied conditions. Here we propose an adaptive chirplet transform (ACT) algorithm coupled with an artificial neural network for robust P300 classification. Comparison of the proposed method with conventional downsampling methods for feature extraction showed an black80% and 73% accuracy respectively when a neural network model was fit to a collection of 16 subject s P300 data. Meanwhile, accuracies of up to 100% could be achieved when the model was trained and validated on subject-specific datasets with the same averaging techniques. This investigation makes it clear that the adaptive chirplet transform holds promise for addressing issues of reliability and robustness in P300 BCI and other related brain signal processing tasks. Here we propose an adaptive chirplet transform (ACT) algorithm coupled with an artificial neural network for robust P300 classification. Comparison of the proposed method with conventional downsampling methods for feature extraction showed an black80% and 73% accuracy respectively when a neural network model was fit to a collection of 16 subject s P300 data. Meanwhile, accuracies of up to 100% could be achieved when the model was trained and validated on subject-specific datasets with the same averaging techniques. This investigation makes it clear that the adaptive chirplet transform holds promise for addressing issues of reliability and robustness in P300 BCI and other related brain signal processing tasks. This investigation makes it clear that the adaptive chirplet transform holds promise for addressing issues of reliability and robustness in P300 BCI and other related brain signal processing tasks.