An Empirical Evaluation of Brain Computer Interface Models from a Pragmatic Perspective

Abstract

Designing Brain Computer Interfaces (BCIs) is a multi-domain task that involves selection of sensing elements to capture brain signals, pre-processing these signals, their segmentation to obtain Regions of Interest (RoI), feature extraction & selection from these RoIs, classification into computer actions, and post-processing tasks. A wide variety of Machine Learning based methods are proposed by researchers, and each of them have their own internal & external configuration & dependency characteristics. For instance, methods like Convolutional Neural Networks (CNN) are highly generic but provide moderate BCI accuracy, while ensemble methods are existing BCI Models, and evaluates them in terms of their contextual nuances, functionality specific advantages, deployment specific limitations, internal & external reconfiguration requirements, and application specific future scopes. Based on this discussion, readers will be able to identify function models for their context-specific use cases. This text also compares these models in terms of qualitative metrics including accuracy levels, precision levels, computational complexity, cost of deployment, and scalability capabilities. Referring this comparison, readers will be able to identify optimum models for their performance-specific use cases. To further simplify the process of BCI Model selection, this text proposes evaluation of a novel BCI Rank Metric (BRM), which combines these parameters in order to identify models that can be used for multiple performance-specific use cases.