Supervised machine learning scheme for tri-axial accelerometer-based fall detector

Fall events can cause trauma, disability and death among older people. Accelerometer-based devices are able to detect falls in controlled environments. The paper presents a computationally low-power approach for feature extraction and supervised clustering for people fall detection by using a 3-axial MEMS wearable accelerometer, managed by an stand-alone PC through ZigBee connection. The paper extends a previous work in which fall events were detected according to a threshold-based scheme. The proposed approach allows to generalize the detection of falls in several practical conditions, after a short period of calibration. The clustering scheme appears invariant to age, weight, height of people and relative positioning area (even in the upper part of the waist), overcoming the drawbacks of well-known threshold-based approaches in which several parameters need to be manually estimated, according to the specific features of the end-user. In order to limit the workload, the specific study on posture analysis has been avoided and a polynomial kernel function is used while maintaining high performance in terms of specificity and sensitivity. The supervised clustering step is achieved by implementing an One-Class Support Vector Machine classifier.