Abstract
Osteoporosis constitutes a significant public health concern necessitating proactive prevention, treatment, and monitoring efforts. Timely identification holds paramount importance in averting fractures and alleviating the overall disease burden. The realm of osteoporosis diagnosis has witnessed a surge in interest in machine learning applications. This burgeoning technology excels at recognizing patterns and forecasting the onset of osteoporosis, paving the way for more efficacious preventive and therapeutic interventions. Smart walkers emerge as valuable tools in this context, serving as data acquisition platforms for datasets tailored to machine learning techniques. These datasets, trained to discern patterns indicative of osteoporosis, play a pivotal role in enhancing diagnostic accuracy. In this study, encompassing 40 participants—20 exhibiting robust health and 20 diagnosed with osteoporosis—data from force sensors embedded in the handlebars of conventional walkers were gathered. A windowing action was used to increase the size of the dataset. The data were normalized, and k-fold cross-validation was applied to assess how well our model performs on untrained data. We used multiple machine learning algorithms to create an accurate model for automatic monitoring of users’ gait, with the Random Forest classifier performing the best with 95.40% accuracy. To achieve the best classification accuracy on the validation dataset, the hyperparameters of the Random Forest classifier were further adjusted on the training data. The results suggest that machine learning-based automatic monitoring of gait parameters could lead to accurate, non-laborious, cost-effective, and efficient diagnostic tools for osteoporosis and other musculoskeletal disorders. Further research is needed to validate these findings.