Predicting fall risk using machine learning and computer vision: Development of a clinical decision support system in nursing

Objective: This study aims to develop a system that can predict fall risk using machine learning and computer vision techniques and support clinical decision-making processes in nursing. Materials and Methods: The research was conducted between December 2024 and May 2025 at the Physical Medicine and Rehabilitation Department of Turgut Özal Medical Centre using a design and development model. The sample consisted of 148 individuals who were at risk of falling and met the research criteria. Participants' data were collected using the Morse Fall Scale, computer vision-based gait analysis, and accelerometer devices. The YOLOv8 model and the Keras platform were used for visual analysis and training neural networks, while Logistic Regression, Boosting, Decision Tree, Random Forest, Naive Bayes, SVM, and KNN algorithms were used for classification. Findings: The developed system was able to classify individuals' fall risk with high accuracy using kinematic data related to walking. A statistically significant relationship was found between the Morse Fall Scale and the system outputs (p<0.05). The Boosting based classifier was found to be more successful than other methods. Conclusion: This study demonstrated that computer vision and machine learning techniques can be used as effective tools for predicting the risk of falling in clinical settings. It was concluded that this system, which can be integrated into nursing practices, can make significant contributions to improving patient safety and preventing complications associated with falling. Keywords: Machine learning, Computer vision, Fall risk, Clinical decision support system, Nursing