Assessing mental workload using portable neurotechnology

What is it about?

Functional near-infrared spectroscopy (fNIRS) is a non-invasive neurotechnology, measuring brain activity, that can be very portable and has shown promise for human-computer interaction. We explore in this work different machine learning approaches for classifying mental workload from fNIRS brain data recordings during computer-based tasks and assess their performance and limitations. In particular, we investigate three different machine learning models: a simple logistic regression, a support vector machine, and a deep learning convolutional neural network. We examine personalised (training the models on each subject individually) and generalised (training the models on multiple subjects at once) approaches, as well as consider different features and ways of labelling the data using subjective mental workload ratings provided by the participants during the tasks.

Featured Image

Photo by Growtika on Unsplash

Photo by Growtika on Unsplash

Why is it important?

Our explorations show that generalised approaches can perform as well as personalised ones and that deep learning can be a suitable approach for relatively small datasets, providing useful insights for scientists aiming at classifying mental workload more accurately in real-time using fNIRS neurotechnology, and building brain-computer interfaces. We also provide practical advice discussing the limitations and data-preparation needs of different machine learning approaches.

Perspectives