Quantitative Assessment of Fall Risk in the Elderly Through Fusion of Millimeter-Wave Radar Imaging and Trajectory Features

The aging process and chronic diseases can lead to functional decline in older adults, particularly with significant decreases in balance ability, which greatly increases the risk of falls. Traditional balance ability assessment methods typically rely on clinical rating scales, which are subjective and prone to the Hawthorne effect and are difficult to implement for continuous daily assessment. In this paper, we propose a quantitative risk assessment system for elderly falls based on the fusion of millimeter-wave(mmWave) radar imaging and trajectory features. Key features such as the center-of-mass trajectory, trajectory offset, and maximum swing diameter are extracted by improving the fusion clustering algorithm. Then, a model such as Random Forest (RF) is applied to conduct correlation analysis on the features, ultimately proving a significant correlation between feature selection and scale scoring. Subsequently, a quantitative assessment model is established with core algorithms such as Support Vector Regression (SVR), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) to verify the effectiveness of the evaluation. The result indicates that the LightGBM model achieved the best performance in assessment compared to other models, with a prediction accuracy score of 93.36%. The experiment has demonstrated that the system can effectively capture the features of gait and evaluate early changes in balance ability decline. The research on this system provides a new technological approach to daily fall risk warnings.