Multidimensional balance assessment and machine learning gait stability prediction in elderly women

To develop and validate a comprehensive balance assessment scale specifically designed for elderly women and construct predictive models for gait stability outcomes using machine learning approaches. A total of 276 community-dwelling elderly women aged 60-80 years participated in this study. A multidimensional balance assessment scale was developed through expert consultation and psychometric evaluation. Three-dimensional gait analysis was conducted to collect biomechanical parameters. Multiple machine learning algorithms including random forest, support vector machine, and gradient boosting were employed to construct gait stability prediction models using balance assessment scores, functional reach distance, single-leg stance duration, tandem walking performance, and demographic variables as predictor inputs. The 12-item balance assessment scale demonstrated excellent reliability (Cronbach's α = 0.89, ICC = 0.88) and validity (CVI = 0.92). Moderate to moderately strong correlations were found between balance scores and gait parameters (r = 0.35-0.67). The ensemble machine learning model achieved superior predictive performance (R² = 0.71, AUC = 0.91) for gait stability classification. External validation confirmed adequate generalization capability (R² = 0.66, indicating a large effect size per Cohen¹). The validated balance assessment scale and machine learning-based prediction models demonstrate significant potential for identifying elderly women at risk for gait instability in community exercise programs and clinical settings.