Early Detection Of Falls In The Elderly Using Frequency Domain Features Of The Acceleration Signals

Health problems in elderly population such as falls increase the economic burden on the health system. It is very important to determine the fall beforehand in order to protect the health of the elderly and reduce the economic burden on the health system. It is very important to determine the fall beforehand in order to protect the health of the elderly and reduce the economic burden on the health system. In order to determine the fall beforehand, the elderly should be regularly checked in the primary health care centers. Therefore, it is a current need to develop a simple and easily applicable system in primary health care centers. In this study, it is aimed to develop a system that can perform this process with a sensor and a short-term recording during an activity. For this, the acceleration signals recorded from the accelerometer in the waist region of the 71-aged person during a one-minute walk were used. The gravitational component was first extracted from the signals obtained from the acceleration sensor, and the power spectral densities were found after filtering and normalization. Later, a total of 87 features were obtained, 29 from each axis. The feature selection process has reduced the amount of features and the classification process has been made using support vector machines. Two different classification models were used in the study and the highest classification accuracy was obtained as 72.6% (AUC = 0.8). The fact that we try to separate the fall and the control group from the data recorded from a sensor during an activity and the use of different power spectrum features that have not been used to solve this problem before in the literature are the points that distinguish our study from the literature.