Threshold Tuning-Based Wearable Sensor Fault Detection for Reliable Medical Monitoring Using Bayesian Network Model

As the medical body sensor network (BSN) is usually resource limited and vulnerable to environmental effects and malicious attacks, faulty sensor data arise inevitably which may result in false alarms, faulty medical diagnosis, and even serious misjudgment. Thus, faulty sensory data should be detected and removed as much as possible before being utilized for medical diagnosis-making. Most available works directly employed fault detection schemes developed in traditional wireless sensor network (WSN) for body sensor fault detection. However, BSNs adopt a very limited number of sensors for vital information collection, lacking the information redundancy provided by densely deployed sensor nodes in traditional WSNs. In light of this, a Bayesian network model-based sensor fault detection scheme is proposed in this paper, which relies on historical training data for establishing the conditional probability distribution of body sensor readings, rather than the redundant information collected from a large number of sensors. Furthermore, the Bayesian network-based scheme enables us to minimize the inaccuracy rate by optimally tuning the threshold for fault detection. Extensive online dataset has been adopted to evaluate the performance of our fault detection scheme, which shows that our scheme possesses a good fault detection accuracy and a low false alarm rate.