Research project: Feature extraction in clinical data
Data mining techniques commonly used in engineering are being applied to continuously monitored human bio-indicators to predict health state.
Data mining techniques commonly used in engineering are being applied to continuously monitored human bio-indicators to predict health state.
It has become increasingly possible to monitor a large number of parameters in both experimental and real-time systems. Often this data is used to determine the health of a system at a given point in time and, increasingly, to predict when the system will degrade or fail. When records are kept of such data from a large number of examples of similar systems it becomes possible to post process the data and classify these examples by their current state and past history to improve predictions of their future state.
The aim of this project is to use computational intelligence techniques, for example clustering and particle swarm optimisation, to characterise and cluster data from free-living studies of human volunteers with chronic disease such as diabetes, metabolic syndrome and obesity. The project is investigating the relationships between features in continually measured parameters (e.g. blood sugar level profiles and physical activity patterns) and indicators of health state (e.g. microvascular function and long-term average blood sugar level) measured in a clinic or laboratory. The aim is to determine if patterns detectable within this data are reliable indicators of the health state of an individual.
Bioengineering and human factors