Fall Detection and Activity Recognition using Hybrid Convolution Neural Network and Extreme Gradient Boosting classifier

The branch of study that focuses on ambient assisted living systems has shown a significant amount of interest in the problem of activity and fall detection. These types of systems make use of a variety of sensing technologies to track human movements and attempt to determine the activity being carried out for the goal of health monitoring as well as other applications. In this regard, in addition to activity identification, fall detection is a very essential role. Falls are a leading cause of injuries and even fatalities, hence it is imperative that falls be detected as soon as possible. This study provides a fall detection and activity identification system that not only takes into account the many activities involved in day-to-day life but also takes into account the detection of falls while taking into consideration the intensity and the direction in which the fall occurred. The data from the Inertial Measurement Unit that is included in the SisFall database is first split into non-overlapping segments that last for three seconds each. Following the appropriate augmentation of the data, exacting the feature with the help of a Convolutional Neural Network, followed by an eXtreme Gradient Boosting (XGB) final step for categorization into the different output groups. The results of the studies demonstrate that the gradient-boosted CNN works far better than previous similar approaches, with an unweighted average recall of 88 percent being achieved.