Evolutionary Fuzzy Mixture Models: Applications in Anaesthesia

R. Folland, M.A. Gongora, E.L. Hines, and D.W. Morgan (UK)


Fuzzy Inference System, Evolutionary Algorithm, Anaesthesia, Diagnosis, Classification


This paper presents a novel application of an Evolutionary Programming based Fuzzy Mixture Model (FMM) system for the determination of a patient's suitability for surgical anaesthesia. Over a three month period 150 patients were examined where blood pressure, heart rate, and arterial oxygenation data was taken, alongside clinical history. Each patient was assessed and their suitability for surgical general anaesthesia (GA) was determined by a consultant anaesthetist and categorised into 3 groups: `suitable for GA', `clearly unsuitable for GA', `possibly suitable for GA (referral...)'. In this work, the FMM rules and mixture membership function parameters for a fuzzy classifier were evolved to produce a discrimination system for the above problem. The evolved system was validated against the classifications made by the consultant and compared to benchmark discrimination systems, namely the Multi Layer Perceptron (MLP) and Radial Basis Function Network (RBFN). In our initial experiments the evolutionary FMM generated a series of mixture membership functions and rule-bases which successfully classified 79.53% of the medical dataset. It over performed the benchmark methods, as the MLP could not successfully learn the discrimination hyperplanes, and the RBFN attained a lower correct classification rate. The Evolutionary FMM was shown to be an attractive mechanism for the computationally-lightweight and robust analysis of this kind of data.

Important Links:

Go Back