AUTHOR=Caywood Matthew S. , Roberts Daniel M. , Colombe Jeffrey B. , Greenwald Hal S. , Weiland Monica Z. 

TITLE=Gaussian Process Regression for Predictive But Interpretable Machine Learning Models: An Example of Predicting Mental Workload across Tasks

JOURNAL=Frontiers in Human Neuroscience

VOLUME=Volume 10 - 2016

YEAR=2017

URL=https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2016.00647

DOI=10.3389/fnhum.2016.00647

ISSN=1662-5161

ABSTRACT=There is increasing interest in real-time brain-computer interfaces (BCIs) for passive monitoring of human cognitive state, including cognitive workload. Too often, however, effective BCIs based on machine learning may function as “black boxes” that are difficult to analyze or interpret.
In an effort toward more interpretable BCIs, we studied a family of N-back working memory tasks using a machine learning model, Gaussian Process Regression (GPR), that was both powerful and amenable to analysis. Participants performed the N-back task with three stimulus variants, auditory-verbal, visual-spatial, and visual-numeric, each at three working memory loads. GPR models were trained and tested on data from all three task variants combined, in an effort to identify a model that could be predictive of mental workload demand regardless of stimulus modality. To provide a comparison for GPR performance, a model was additionally trained using multiple linear regression (MLR).The model was effective when trained on individual human participant data, resulting in an average standardized mean squared error (SMSE) between true and predicted n-back levels of 0.44. In comparison, the MLR model using the same data resulted in an average SMSE of 0.55.  
We demonstrate how GPR can be used to identify which EEG features are highly influential in predictions of cognitive workload in an individual participant.  A fraction of EEG features account for the predictive power, and only the top 25% of features were sufficient for near for maximum accuracy. Subsets of features identified by linear models (ANOVA) were not as efficient as subsets identified by GPR. Similarly, multiple linear regression was less effective than GPR for accurately predicting task load.
This raises the possibility of real-time BCIs that are simpler (requiring fewer model features) but that still capture all the neurophysiological information needed to achieve high predictive accuracy. This study sought to establish effectiveness and interpretability for a rapidly trainable, real-time, passive BCI for cognitive monitoring in individuals, without regard for specific theories of neural function. Potential valuable applications beyond the scope of the present study include characterizing the neural basis of workloading or other cognitive states, in individual participants and across groups of human participants.