Predicting the distance between the center of mass and center of pressure during walking in the young and elder population based on ground reaction forces using artificial neural network.

INTRODUCTION: Assessing the interaction between the body's center of mass (COM) and center of pressure (COP) during gait provides valuable insights into dynamic balance control. However, traditional motion capture methods are costly and impractical in clinical settings, highlighting the need for accessible alternatives. This study utilizes force plate data to explore the use of ANNs in predicting the motion of the COM relative to the COP during walking. MATERIALS AND METHODS: Eighty-four healthy adults (50 young, 34 older) performed gait trials on a walkway with motion capture and force plate data collection. Ground reaction forces (GRFs) and COM-COP distances (dCOM) were computed for the stance phase. Multi-layer perceptron (MLP) artificial neural network (ANN) models were developed to predict dCOM in the anteroposterior (AP) and mediolateral (ML) directions as well as COM height based on GRFs and anthropometric inputs. Model performance was evaluated using root mean squared error (RMSE) and coefficient of determination (R2). RESULTS: ANN models achieved high predictive accuracy with optimal hidden layer sizes of 5 (AP, COM height) and 9 (ML). RMSE values ranged from 0.89 % to 7.77 %LL, with R2 values between 0.74 and 0.92. Significant age-related differences were observed in ML direction and COM height (p < 0.05), but not in AP. CONCLUSION: This study highlights ANN models as accurate, low-cost tools for dynamic balance assessment, reliably predicting dCOM across directions. The consistency between calculated and predicted group differences supports their application in fall risk evaluation and personalized intervention planning for older adult populations.