A Multi-Scale Similarity Network Approach to Mobility Profile Detection

Population-level mobility is increasingly studied using wearable and urban audit data, yet most analyses rely on regression or centroid-based clustering that over-look structural similarity patterns. As populations age globally, understanding how activity variability is embedded within demographic and environmental configurations becomes critical for interpreting mobility inequalities. We propose a multi-scale similarity network framework that represents countries and cities as nodes in standardized multidimensional feature space and connects them via k-nearest-neighbor cosine similarity. At the country level, we integrate step inequality, gender gaps, urbanization, and HALE60 to examine structural profiles of mobility and healthy ageing context. At the city level, we apply the same framework to microscale walkability indicators to evaluate whether profile structure generalizes across analytical scales. Community detection and centrality analysis reveal stable similarity profiles and bridge entities not recoverable through partition-based clustering alone. Robustness evaluation using Adjusted Rand Index confirms profile stability under parameter perturbation. The results demonstrate that similarity-based network modeling provides a scale-agnostic structural lens for comparative mobil-ity research and supports policy interpretation in the context of age-supportive environments.