Anomaly Detection in Health Data Based on Deep Learning

Anomaly detection in health data means finding out abnormal human status automatically. In this paper, we propose a scheme which can be used for building monitoring system to promote quality of independent living and reduce the consequences of falls and diseases for elderly. We choose non-contact sensors to collect health data and build system using deep learning algorithms. This scheme includes two approaches, approach based on raw data which aims at abnormal activities and approach based on spectrogram which aims at abnormal status. Convolutional neural network is used to classify activities that predicted by support vector machine later, and recurrent neural network is used to predict signals directly. Through experiments, we evaluate performance of the scheme which proves it can solve the task successfully. Based on this result, it can be expected that our scheme will be utilized in reality.