Evaluation of the autonomic nervous system for fall detection

Studies show that the proportion of elderly will reach 30% of the total population by 2050 in developed countries, such as France. The elderly live generally alone, thus many health problems related to age are under reported. Falling is one of these problems and several devices have been developed recently, based on accelerometers, in order to detect it and alert carers. In order to improve the detection success of these devices, we propose quantifying autonomic nervous system activity (ANS) using a wearable ambulatory device developed for this purpose. We studied the A.N.S's response on 7 adult subjects during simulated falls and standing-lying transitions. We implemented a classification method using the Support Vector Machine in order to classify these two situations using measured heart rate variability and electrodermal response. Good results (sensibility =70.37%, specificity =80%, positive predictor=73.8%) were obtained using a Polynomial kernel (p = 5) for the support vector machine implementation.