AI-based wearable system for fall risk prediction in older adults using sEMG and plantar pressure data

Falls among older adults pose a major healthcare and social burden, making early identification of high-risk individuals essential for prevention. This study presents a portable, non-invasive AI-based wearable system that predicts fall risk using surface electromyography (sEMG) and plantar-pressure measurements collected during overground walking. sEMG electrodes were placed bilaterally over eight key lower-limb muscles—tibialis anterior, peroneus longus, medial and lateral gastrocnemius, rectus femoris, vastus medialis, vastus lateralis, and biceps femoris—while pressure insoles captured loading at eight anatomical foot regions. Ninety-four older adults (mean age 69.6 <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="m1"> <mml:mrow> <mml:mo>±</mml:mo> </mml:mrow> </mml:math> 10.0 years; 57 females), including 57 non-fallers and 37 individuals who met ICD-10 diagnostic criteria for “propensity to fall,” participated in the modeling study. The signals from both devices were streamed wirelessly to a central acquisition unit for synchronized processing. Extracted features included muscle activation contribution, mean frequency, mean power frequency, and cumulative plantar-pressure impulses. These features served as model input. To reduce data dimensionality, Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) were applied. PCA retained a variance structure, whereas LDA maximized class separability. Three machine-learning classifiers—Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGB)—were trained using Leave-One-Out Cross-Validation. LDA substantially improved performance across all models, with LDA + SVM achieving the highest accuracy (0.88), precision (0.92), recall (0.85), and F1-score (0.87). An independent clinical validation study involving ten additional older adults demonstrated that LDA-based models generalized well beyond the original dataset. Compared with existing fall-detection or multimodal EMG-based systems that focus on simulated falls, young participants, or non-portable laboratory equipment, the proposed framework enables physiologically interpretable, clinically deployable fall-risk prediction during natural gait. These findings highlight the promise of dual-modality wearable sensing for proactive fall prevention in geriatric populations.