An application of generalised simulated annealing towards the simultaneous modelling and clustering of glaucoma

Optimisation methods are widely used in complex data analysis, and as such, there is a need to develop techniques that can explore huge search spaces in an efficient and effective manner. Generalised simulated annealing is a continuous optimisation method which is an advanced version of the commonly used simulated annealing technique. The method is designed to search for the global optimum solution and avoid being trapped in local optima. This paper presents an application of a specially adapted generalised simulated annealing algorithm applied to a discrete problem, namely simultaneous modelling and clustering of visual field data. Visual field data is commonly used in managing glaucoma, a disease which is the second largest cause of blindness in the developing world. The simultaneous modelling and clustering is a model based clustering technique aimed at finding the best grouping of visual field data based upon prediction accuracy. The results using our tailored optimisation method show improvements in prediction accuracy and our proposed method appears to have an efficient search in terms of convergence point compared to traditional techniques. Our method is also tested on synthetic data and the results verify that generalised simulated annealing locates the optimal clusters efficiently as well as improving prediction accuracy.