Wearable Sensor-Based Machine Learning in Gait Pattern Analysis for Fall Prediction Systems: An Investigative Research Study

Falls are the primary cause of poor quality of life, morbidity, and death in older persons, whether they have neurological disorders. It is possible to detect those who are at risk of falling in the future by using machine learning models on the results of gait analysis. This study sought to ascertain how various data pre-processing techniques affected the ability of machine learning models to distinguish neurological patients who have fallen from those who have not to evaluate fall risk in the future. Fallers compared to non-fallers walked more slowly, taking longer and shorter steps. The overall accuracy of the model was between 47% and 98%, with a sensitivity of 44-98.9% and a specificity of 45-97.8%. The route signature approach pre-processed the data used to train a random forest classifier, which produced optimum classification accuracy of 97.8% with 98.7% sensitivity and 97.8% specificity. Preprocessing of the data has a direct impact on the precision of machine learning models used to classify fallers. Clinical decision-making may be supported, and proactive fall risk assessment can be carried out with the use of gait analysis and trained machine learning models.