The rapid evolution of wearable sensors and mobile edge computingEdge computing has heralded a new era in health and safety monitoring, enabling continuous, unobtrusive surveillance of human motion for at-risk populations. Human activity recognition (HARHuman Activity Recognition (HAR)) systems classify routine behaviors—such as walking, sitting, standing, and lying down—by interpreting streams of accelerometerAccelerometer and gyroscopeGyroscope data. Fall detectionFall detection (FD), a critical subset of HARHuman Activity Recognition (HAR), focuses on identifying abrupt, potentially injurious events that disproportionately affect elderly individuals and patients with movement disorders. Traditional FD methods typically apply simple threshold rules to the signal vector magnitude (SVMSupport Vector Machine (SVM)) of triaxial acceleration, triggering alarms whenever SVMSupport Vector Machine (SVM) exceeds preset limits. However, such rule-based solutions often misinterpret fast sit-to-stand motions, stumbles, or device knocks as genuine falls, resulting in high false alarmFalse-alarm rates and caregiver “alarm fatigueAlarm-fatigue.” In response, modern approaches integrate machine learningMachine Learning (ML) (ML) to extract rich time–frequency features and learn complex decision boundaries that more accurately discriminate between routine movements and dangerous falls, improving both sensitivity and specificity in real-world deployments.