Psychophysiological dynamics of neurodiverse and neurotypical dyads during remote collaboration: a machine learning approach

Abstract Objective: We aimed to explore differences in physiological dynamics of neurodiverse (autistic–nonautistic) and neurotypical (nonautistic–nonautistic) adult dyads during remote collaboration. Method: 3 autistic adult (AA) and 15 nonautistic adult (NA) participants were paired into 9 age- and gender-matched dyads (3 AA-NA, 6 NA-NA) who collaborated on a simulated remote work task while synchronized interbeat interval time-series data were collected from both participants. We used 2 Hidden Markov Models (HMM) to analyze these data: (1) a group HMM trained on data from all dyads and (2) separate HMMs trained on data from each dyad. Models were evaluated via log likelihood (LL) and perplexity and interpreted via transition probabilities, state occupancy, entropy, and dwell-time. Results: Across both models, AA-NA dyads exhibited better model fit (greater LL, lower perplexity) compared to NA-NA dyads. AA-NA dyads were more likely to occupy specific hidden states for longer periods throughout the task compared to NA-NA dyads. NA-NA dyads followed a less-predictable pattern between states marked by frequent transitions and higher entropy. Conclusions: Compared to NA-NA dyads, AA-NA dyads expressed fewer transitions and longer state occupancy in hidden states, suggesting a more and persistent autonomic state throughout the collaborative task.