In a study published in the journal Information Systems Research, Shuo Yu and his collaborators from Texas Tech University introduced a novel generative machine learning model designed to identify instability signals preceding a fall. The ultimate goal of this model is to integrate with fall detection devices like anti-fall airbag vests or medical alert systems to proactively prevent falls, thereby minimizing injuries, enhancing emergency responses, and reducing healthcare expenses.
“You can treat this as a kind of AI,” explained Yu, who serves as the Wetherbe Professor of Management Information Systems at the Jerry S. Rawls College of Business. “It detects your moving status and predicts if there’s going to be a fall. It can help mitigate injuries automatically.”
Yu and his team leveraged two publicly available datasets containing data from wearable motion-sensor devices monitoring nearly 2,000 falls. By meticulously labeling and categorizing data points, they identified three key hidden states of a fall: collapse, impact, and inactivity.
Rather than relying on traditional rule-based models, the researchers developed a sophisticated model known as a hidden Markov model with generative adversarial network (HMM-GAN). This model combines the statistical power of HMM for analyzing sequences over time with the generative capabilities of GAN for creating realistic data.
The HMM-GAN model excelled in predicting falls across various experiments, outperforming existing frameworks and demonstrating faster response times. This innovation holds immense potential for senior care, offering increased peace of mind to families and significant economic benefits to healthcare facilities.
Yu envisions a future where AI-powered devices like the one he developed can significantly improve health outcomes and enhance the quality of life for individuals, especially senior citizens. The positive results of this study serve as a promising proof-of-concept for future research and product development in related fields.
