Falling is a prevalent and hazardous event that can lead to severe injuries, such as limb fractures or spinal damage, especially for elderly individuals in hospital care. In this study, we aim to develop a machine-learning-based system for effective fall detection and prompt intervention. We applied deep learning techniques, particularly the AlphaPose + Spatial Temporal Graph Convolutional Network (ST-GCN) model, to enhance human activity recognition and behavior analysis. These advanced machine learning models allow for the real-time monitoring of fall events by accurately identifying abnormal movements and behaviors associated with falls. In this study, we employed a web camera as a sensor to capture the human pose, and the AI-powered system achieved an accuracy rate exceeding 96% in training results, showcasing its robustness in detecting falls. Upon detection, the system sends immediate alerts via communication software, ensuring timely notifications to healthcare providers or family members. This machine learning approach significantly improves the safety of elderly individuals by reducing response time and minimizing the risk of fall-related injuries.