Health Related Time Series Mining And Application

With economic development and improvement of living standards, a variety ofchronic diseases and sub-health groups is rising year by year. People gradually realizedthe importance of health, while the current health care system can’t meet the growingdemand for health management. Based on the e-health project of our laboratory, Iconducted a comprehensive survey on health management system architecture andhealth-related data mining. Regarding to the application problems in the existing miningmethods I made in-depth study and found there is need to improve it.First, I surveyed the mainstream building solutions of health management platform,which involves the hardware, wireless transmission and other core issues. Some of theexisting systems were compared and analyzed, providing the subsequent projectdevelopment some reference basis. The article also introduced data mining applicationsin healthcare with the management of diabetes as an example and established a datamining solution for continuous glucose monitoring：the principle idea is to handle shortterm analysis and data compression using gateway devices and then transmit thecompressed data to server for long term storage and analysis.Regarding to continuous glucose monitoring related data transmittal and miningrequirements, we made a comprehensive survey of latest research in time series mining.Based on the survey we decided to use PLA (piecewise linear approximation) as ourmethod for data compression and designed a new version of DTW (Dynamic TimeWarping) based on piecewise linear approximation, named TDTW (Trended DynamicTime Warping). A comprehensive experimental verification shows that TDTW has acertain practical value.According to the shortcomings of TDTW showed in real applications, weconducted a series of strategies and achieved specific improvement.Finally, we provide a summary of the main content of the article, and prospectsubsequent progress of the project and relevant academic research.