A neurological approach to classify trust through EEG signals using machine learning techniques

Trust is an important factor in any teamwork, including human-human teaming as well as human-machine teaming; therefore, measuring and calibrating trust is important. One of the quantitative ways to measure and calibrate trust is by observing neural signals. Most of the studies, so far, have been conducted for a specific condition, that is- to understand trust in a specific complex background, e.g., in different trust games or some kind of simulation environment. These results are difficult to infer in a general trust scenario. The quantitative trust measurement in the basic context is yet to be delved into. In our study, we aim to calibrate trust quantitatively by analyzing the neural signal to address this perspective. We chose to perform a ‘word elicitation’ experiment to keep the context basic and aligned with sociological contrast, where a list of words associated with trust and mistrust has been shown to the participants and the participants are expected to think about the trust/mistrust situation corresponding to those words. During this process, neural signals have been collected as Electroencephalography data, and the power spectrum density analysis is done in the frequency domain. In this study, we performed classification analysis with the two labels ‘trust’ and ‘mistrust’ on the power spectrum density of alpha, beta, and gamma frequency range. We developed a machine learning model and performed several machine learning classification algorithms on the model to find out the best algorithm with this dataset. We achieved 72% average accuracy with the Decision Tree classification algorithm. In the future, we will extend this study to test this model in a trust/mistrust situation where the human operator makes decisions based on trust or mistrust while teaming with an autonomous system.