Nonparametric Supervised Learning for Enhancing BCI Performance

Abstract

Improving the function and reliability of brain-computer interfaces (BCIs) is an important factor in facilitating their usage to impact human wellbeing. This research developed a BCI system that utilized nonparametric feature extraction and dimension reduction and supervised learning as a framework for improving accuracy. A BCI experiment using steady-state visually evoked potentials (SSVEP) was conducted as a test basis for our framework. Typical unsupervised learning BCI techniques were tested and found to be improved when harmonic frequencies were included as inputs. Nonparametric weighted feature extraction (NWFE) and physiologically relevant input features were found to improve supervised learning classifiers in our BCI framework, which could outperform the comparable unsupervised methods. This framework presents a novel basis for enhancing BCIs which take into account known physiological information and NWFE to perform better.