The disease combination appearance probability matrix for learning-based diagnosis in u-health applications

Abstract

U-Healthcare service is a new service which integrates various ubiquitous technologies such as portable vital sign sensors and high-speed communication infrastructure based on traditional healthcare services. Selecting an appropriate decision support method is one of issues in development of u-health applications. However, it is not easy to adopt the traditional machine learning algorithms to the u-health applications without any modifications because the u-healthcare is somewhat different from the traditional healthcare. This paper proposes two requirements of the method in u-healthcare and introduces a learning-based framework for meeting all requirements. The framework uses a matrix in order to count appearance frequencies support incremental learning by adding counts into the existing value and devise an impact matrix to give analyzing information of a diagnosis. We conducted three experiments to validate our framework. Through the results of three experiments, we could give a brief guideline to the framework designer, and the performance of the framework was confirmed by comparing with SVM. Also, we use three criteria of incremental learning to validate our method that supports incremental learning. There are some limitations such that this framework requires the most number of ranges as possible and people may be not received same quality of service until stable condition, but we can solve these limitations after constructing u-health environment fully. We hope that many u-health applications apply our framework to develop new diagnosis applications.