Balance and fall-risk assessment in older adults using wearable plantar pressure and semi-supervised learning

Falls are a major public health concern among older adults, leading to disability, reduced independence, and high healthcare costs. Conventional balance assessments such as the Berg Balance Scale are limited by subjectivity, time requirements, and dependence on trained evaluators, creating barriers for large-scale community application. To address these challenges, we developed an intelligent footwear system combined with a semi-supervised learning framework to objectively predict Berg Balance Scale scores and assess fall risk. In a study of 136 older adults aged 60-90, plantar pressure signals from smart insoles with eight sensors per foot were collected, and 156 biomechanical features were extracted. A multi-model error consistency approach was applied to mitigate label noise, and feature selection identified ten interpretable predictors related to pressure duration, peak intensity, and inter-limb symmetry. The model achieved root mean square errors of 3.99 in validation and 3.13 in an independent test group. This wearable-based, interpretable, and scalable approach provides a practical solution for early detection of fall risk, enabling timely community interventions and supporting healthy aging strategies in public health.