An Explainable Deep Learning Model for Vision-based Human Fall Detection System

Human Fall is one of the major life-risking problems among elderly people. Accidental falls among elder persons may lead to many irreversible disabilities or even deaths. Fall detection has become a critical research problem in the Healthcare domain, which needs more reliable and efficient solutions to intelligently classify fall activities. With the development of the Internet of Things, such as wearable sensors, ambient sensors, and cameras, monitoring elderly people continuously has become feasible. The proposed work ensures Fall detection through Deep Learning classification using a Convolutional Neural Network with three hidden layers. The performance of the Deep Learning model is evaluated with the Fall Detection Dataset and the classification results have achieved an accuracy of 96.5%. However, the intelligent classification methods need to be trustworthy to be able to be accepted by healthcare professionals. Since the Deep Learning models are ’black-box’, the process behind the classification is not known and accuracy alone will not be sufficient for evaluating the performance of the model. Hence, in this paper, we also propose an Explainable AI model, called LIME to interpret the classification of fall activity. the results of LIME show the feature responsible for prediction by marking the boundaries of the input image.