Fault diagnosis of body sensor networks using hidden Markov model

In this paper, we focus on medical body sensor networks collecting physiological signs to monitor the health of patients. We propose a Hidden Markov Model (HMM) based method for fault diagnosis of measured data transmitted from sensors. We firstly verify the Markov property of temporal data sequences from medical databases. Then we improve the Baum-Welch algorithm at two aspects to estimate parameters of HMMs by history training data, and use the Viterbi algorithm to determine whether the new sensor reading is faulty. Finally, we do experiments on both real and synthetic medical datasets to study the performance of the fault diagnosis method. The result shows that the proposed approach possesses a good detection accuracy with a low false alarm rate.