Estimación de la velocidad de marcha en adultos mayores mediante técnicas de aprendizaje automático y regresión lineal
Loading...
Date
2026-04-05
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Tecnológica Centroamericana UNITEC
Abstract
La fragilidad y el deterioro funcional aumentan significativamente el riesgo de dependencia, caídas y disminución de la calidad de vida en adultos mayores. La velocidad de la marcha es un indicador clave del deterioro funcional y una herramienta valiosa para la detección temprana de la fragilidad.
Este estudio tiene como objetivo desarrollar y comparar algoritmos basados en sensores inerciales para la estimación de la velocidad de la marcha, utilizando técnicas de aprendizaje automático y regresión lineal. Se analizaron datos de los proyectos ActiveUP y MOTIVA en un estudio transversal para validar modelos predictivos en una población geriátrica.
El rendimiento de los modelos se ha analizado mediante la validez (correlación), la concordancia (SEM y Bland-Altman) y la fiabilidad (ICC). Se han explorado diferentes modelos de regresión para el último paso del algoritmo de Mueller.
El modelo SVM mostró el mejor equilibrio en concordancia (SEM = 0.07, MDD = 0.18), mientras que el modelo RF presentó la mayor validez (R = 0.98, IC 95%: 0.98 - 0.99).
Estos hallazgos resaltan el potencial de la tecnología de sensores portátiles y el aprendizaje automático para una estimación fiable de la marcha, apoyando la detección temprana de la fragilidad y fomentando estrategias de atención preventiva en adultos mayores.
Sin embargo, el tamaño limitado de la muestra subraya la necesidad de realizar más estudios para garantizar la aplicabilidad clínica y una generalización más amplia.
Frailty and functional decline significantly increase the risk of dependence, falls, and reduced quality of life in older adults. Gait speed is a critical indicator of functional impairment and a valuable tool for the early detection of frailty. This study aims to develop and compare inertial sensor-based algorithms for gait speed estimation, employing machine learning and linear regression techniques. Data from the ActiveUP and MOTIVA projects were analyzed in a cross-sectional study to validate predictive models in a geriatric population. Model performance was assessed through correlation analysis, intra-class correlation (ICC), and Bland-Altman analysis. Four models were evaluated, including the adapted Mueller algorithm, linear regression (MOTIVA), support vector machines (SVM), and random forests (RF). The SVM model demonstrated the best balance between validity and reliability (SEM = 0.07, MDD = 0.18), while the RF model showed the highest validity (R = 0.98, 95% CI: 0.98–0.99). These findings highlight the potential of wearable sensor technology and machine learning for reliable gait estimation, supporting the early detection of frailty and promoting preventive care strategies in older adults. However, the limited sample size highlights the need for further studies to ensure clinical applicability and broader generalization.
Frailty and functional decline significantly increase the risk of dependence, falls, and reduced quality of life in older adults. Gait speed is a critical indicator of functional impairment and a valuable tool for the early detection of frailty. This study aims to develop and compare inertial sensor-based algorithms for gait speed estimation, employing machine learning and linear regression techniques. Data from the ActiveUP and MOTIVA projects were analyzed in a cross-sectional study to validate predictive models in a geriatric population. Model performance was assessed through correlation analysis, intra-class correlation (ICC), and Bland-Altman analysis. Four models were evaluated, including the adapted Mueller algorithm, linear regression (MOTIVA), support vector machines (SVM), and random forests (RF). The SVM model demonstrated the best balance between validity and reliability (SEM = 0.07, MDD = 0.18), while the RF model showed the highest validity (R = 0.98, 95% CI: 0.98–0.99). These findings highlight the potential of wearable sensor technology and machine learning for reliable gait estimation, supporting the early detection of frailty and promoting preventive care strategies in older adults. However, the limited sample size highlights the need for further studies to ensure clinical applicability and broader generalization.
Keywords
Envejecimiento Saludable, Estimación de la Velocidad de la Marcha, IMUs, Regresión Lineal, Aprendizaje Automático