Bridging Traditional and AI-Driven Fall Risk Assessment: A Flexible Physiological Patch for Elderly Safety and Independence

Accurate assessment of physiological signals during functional mobility tasks is critical for detecting subtle changes in postural stability, particularly in elderly populations at risk of falls. In this work, we propose a convolutional neural network (CNN) trained on wavelet-filtered surrogate features that summarize ECG signal power within specific frequency ranges. Instead of employing a traditional time–frequency spectrum, wavelet filtering was applied to extract frequency-band-restricted motion signatures, producing surrogate IMU-like features from a single ECG channel. These features were then used as input to the CNN for classifying baseline versus weighted conditions. All validation in this study focused on the sit-to-stand transition (“Arises from the chair”) from the Tinetti Score framework, a task widely recognized as a sensitive indicator of lower-limb strength and balance control. To provide a reliable estimate of generalization performance, we employed repeated stratified hold-out validation (20 runs) across 176 samples (88 baseline and 88 weighted). The proposed CNN achieved an average accuracy of 80.5%±9.0, precision of 79.7%±10.8, recall of 88.8%±9.7, specificity of 71.3%±22.9, F1-score of 83.0%±6.0, and a ROC–AUC of 0.891±0.089. These results confirm the feasibility of deriving motion surrogates directly from ECG and demonstrate that reliable classification can be achieved without continuous IMU streaming. It should be noted, however, that the current dataset was collected from healthy adults, with aging effects simulated by weighted conditions rather than real elderly participants. Future studies will therefore validate the system on elderly populations with age-related physiological complexities. Beyond classification accuracy, the approach also offers substantial system-level advantages. By eliminating continuous IMU transmission and relying only on a single ECG channel, the data rate was reduced by nearly 90%, leading to lower power consumption and improved wearability for long-term monitoring. This highlights the potential of ECG-only sensing as an efficient, scalable solution for elderly care, fall risk assessment, and other applications where energy efficiency, comfort, and privacy are critical.