| With the improvement of living standards, people are more and more concerned about their health condition. Pervasive computing and information technology can provide people with universal health care and medical services timely. Pervasive health monitoring, which is designed for the elderly and high-risk patients with a variety of chronic diseases, can make full use of the existing medical resources and greatly reduce unnecessary costs. In order to improve the quality of service, clinicians need to analyze plentiful data collected by medical information system. In this way, useful knowledge can be acquired and wise decision will be made. However, since medical diagnosis involves many uncertainties, making medical decisions with ubiquitous health monitoring system is full of difficulties. How to represent this uncertainty clearly has become a hot and difficult spot in pervasive health monitoring applications.In this paper, taking the special nature of medical data into account, we use Bayesian Networks to process the heart disease data and build an intelligent probabilistic model to diagnose the disease. First we propose a method to learn the sequence of network nodes from sample data set. This method overcomes the limitation of traditional algorithms, which require experts of medical field to give the sequence of network nodes. In order to shorten the modeling time, a parallel optimization technique is then adopted to accelerate the establishment of diagnosis model over large amounts of data. After that, we give a detailed description of the entire modeling process, including data preprocessing. For the defect of single classifier, this paper presents a classifier ensemble method in which both accuracy and diversity of the base classifier are considered. So that the generalization ability of classifier can be enhanced. Finally, experiments show that the several proposed methods can improve the accuracy of the modeling and shorten the modeling time. |
PDF Full Text Request