Modelling and Forecasting of Glaucomatous Visual Fields Using Genetic Algorithms

The prediction of visual field deterioration in patients who are suffering from normal tension glaucoma plays an important role in the management of the disease. The Vector Auto-Regressive (VAR) process appears to be an appropriate way of modelling the multivariate time series data from the visual fields. However, standard parameterisation techniques such as the Yule-Walker equations for building a VAR model place a restriction on the minimum length of time series observations. In this paper genetic algorithms are suggested as a way of finding the order and estimating the parameters for the VAR process. To evaluate the effectiveness of this approach, the VAR process in S-Plus, the Holt-Winters forecasting method, and a pure noise model are applied to the same set of visual field data.