Predicting Dyadic Synchrony: A Theory-Driven Machine Learning Approach.

Dyadic synchrony, defined as the coordination of emotional and behavioral signals between mother and child, is a central mechanism supporting early socioemotional development. However, predicting its emergence remains challenging because relational and psychological influences may combine in complex, context-sensitive ways that are not easily captured by traditional linear models. The present study applied a theory-guided machine learning approach to examine predictors of observed synchrony in 204 mother-child dyads (mean maternal age = 33.7 years, 89% White, 52.6%, boys mean children's age = 11.71 months), drawn from the PEACE (Perinatal Experiences and COVID-19 Effects) Study, with data collected between November 2021 and August 2022. Dyadic synchrony and affective behaviors were coded from free-play interactions using the Coding Interactive Behavior system, and psychological predictors included maternal anxiety, resilience, parenting stress, and observed affective expressions. Random Forest models were compared with linear regression, and model interpretation was supported using SHAP and ALE techniques. Linear regression showed slightly stronger predictive performance, whereas Random Forest revealed potentially meaningful non-linear and interaction-based patterns. In particular, synchrony tended to decrease under conditions reflecting imbalance between parenting stress and child positive affect, and affective mismatch showed a non-monotonic association with synchrony. These findings highlight the context-dependent nature of dyadic coordination and suggest that synchrony may emerge from configurations of stress, affect, and emotional alignment rather than from simple additive effects. More broadly, the study illustrates how interpretable machine learning can complement traditional statistical models to explore complex relational processes in early development.