Quantification and Visualization of Interpersonal Synchrony Using Wearable Sensors: A Case Study on Autistic and...

Interpersonal synchrony (IS), a key indicator of social interactions, is traditionally assessed through video data and manual coding methods, a process that is time-consuming and subjective. This study presents an automated sensor-based framework for quantifying and visualizing IS using time series data collected from wearable sensors, demonstrated through a case study of interactions between autistic and neurotypical children in classroom settings. We evaluated time series similarity measures, including Cross-correlation (CC), Dynamic Time Warping (DTW), and Cross-Wavelet Analysis (XWA), as features for machine learning (ML) models that classify interaction levels, where ground truth labels are derived from video-coded motor coordination as a behavioral proxy for IS. Results show that these similarity-based features outperform conventional statistical features in distinguishing high and low IS using ensemble classifiers. We further compare two approaches for identifying pseudosynchrony: a surrogate data analysis for threshold estimation and a supervised learning approach for direct prediction, providing a systematic evaluation of their methodological trade-offs that has been largely overlooked in prior synchrony research. The developed visualization tools enable dynamic tracking of interaction patterns while filtering out pseudosynchrony. The proposed workflow offers a scalable, objective, and reproducible alternative to manual coding, addressing a key gap in the current literature and supporting broader applications in social, developmental, and rehabilitation research.