An advanced integrated sensor-based method for fall risk assessment in rehabilitation setting

Abstract

Falls are the most common preventable adverse events in hospitals and are strongly linked to movement-related disorders. Conducting fall risk assessments and implementing personalized interventions for older adults in sports rehabilitation settings can significantly reduce fall incidence. Sensor-based techniques and machine learning models offer new opportunities for measuring gait and balance in a more sophisticated way to enhance fall risk assessments. This study aims to develop an integrated sensor method to provide continuous and effective fall risk assessments for older adults in rehabilitation settings. A joint feature extraction scheme based on integrated sensors was proposed, including temporal features from inertial measurement unit (IMU) signals and spatial features from depth camera data during 3-m timed up and go (TUG) test. A set of classifiers, including support vector machine (SVM), logistic regression (LR), k-nearest neighbors (KNNs), random forest (RF), and eXtreme gradient boosting (XGBoost), were used in conjunction with a feature selection strategy to facilitate developing the predictive models for fall risk assessment. We conducted validation experiments using real-world data for comprehensive comparative analysis. The results demonstrate that our integrated approach achieves superior classification performance (AUC: 0.8633–0.9586). These findings suggest that the complementary features from sensors have advantages in bridging information gaps, reducing missed diagnoses, and assisting clinicians in early fall risk identification. The proposed method shows significant potential to deliver comprehensive fall risk assessments for older adults in rehabilitation settings.