Classification of Fallers Versus Non-Fallers From Gait Analysis of Older Adults using Long Short-Term Memory (LSTM)
Gait analysis using a lightweight body-worn sensor is a convenient and inexpensive method that potentially allows for easy and quantitative assessment of fall risks for the geriatric population. In this study, older adults who each worn a gait belt on the lower back were evaluated using data obtained from lightweight sensors including a triaxial accelerometer and a triaxial gyroscope. Algorithms such as artificial neural network (ANN), support vector machine (SVM) and Long Short-Term Memory (LSTM) were used to classify fallers vs. non-fallers. The accuracy obtained by focusing on the temporal change of gait patterns using LSTM is 72.72% for the 1-min lab walk and 81.81% for the 3-day recording of daily movement, which are comparable to state-of-the-art prior arts reported in the literature. These results suggest that long-term monitoring on the temporal changes in gait behaviors can be a rather effective and novel technique for quantitively evaluating fall risks on the community-dwelling older adults performing their daily living activities.
Keywords - Artificial Neural Network (ANN); Support Vector Machine (SVM); Long Short-Termmemory (LSTM).