Electromyography and dynamometry in the prediction of risk of falls in the elderly using machine learning tools

The aging process affects mechanisms for maintaining physical integrity. The assessment of the risk of falls is routine in the services of assistance to the elderly, but subjective and time-consuming, so that the automation of the process is desirable as a supporting tool. In this regard, the aim of this study is to carry out a comprehensive evaluation study on the use of machine learning techniques as an instrument to predict the Berg Balance Scale (BBS) score, using different sets of electromyographic and dynamometric data collected during a voluntary isometric contraction. Thirty participants were evaluated with the BBS and with electromyography and dynamometry of the vastus lateralis, biceps femoris, lateral gastrocnemius and tibialis anterior muscles during maximal isometric voluntary contractions. After pre-processing the dataset, the features were selected through principal components analysis (PCA), correlation-based function select (CFS) and relief-F to then be applied to the multilayer perceptron (MLP), random forest (RF), random tree (RT), k-nearest neighbors (KNN), Multiple Linear Regression (MLR), and least-squares support vector regression (LS-SVR). From the fitted regression models, our ultimate goal is to infer which selected features correlate most with the risk of falling for elderly people and how those features connect themselves to certain groups of muscles. In this regard, the features extracted from myoelectric signals proved to be more effective for use in predicting the risk of falls in the elderly in relation to the force-related signal. The proposed models obtained good results for predicting the BBS score and classifying the risk of falls.