Publication

Abstract : Falls are exceptional activities that put one’s health in danger. To limit the effect of falls, it is necessary to build fall detection and prevention systems. The goal of emerging technology is to create such systems that will improve people’s quality of life, especially for the elderly. To limit the danger of injury, a fall detection system detects the fall and generates an assistance signal. The suggested system detects falls by classifying various behaviors as fall or non-fall activities and alerting those who are affected in the event of an emergency. To calculate characteristics, the dataset SisFall is used, which contains a variety of actions performed by numerous people. The machine learning methods XGBoost and LightGBM are used to identify falls based on calculated characteristics. Using the XGBoost algorithm, the system achieves ROC-AUC scores of up to 97.64%. Our proposed solution is based on a transformer model, which is then tailored to produce the best outcomes, with an accuracy of approximately 95.7%.