Contribution to Skin Cancer Prevention in South Africa: Modelling the UV Index Utilizing Imprecise Data

Abstract

South Africa has a high incidence of skin cancer and eye disorders because of the high number of sunshine hours per day. The ultraviolet (UV) index (UVI) provides a factual measure of the UV irradiance including biological effects. It is considered extremely important when gauging ultraviolet doses. A survey conducted across South Africa during January 1999 provided data records that contain nine independent variables for UVI inference purposes. This set of data includes cloud cover and other subjectively observed variables such as turbidity. The data set was recorded at 272 locations. Modelling the UVI by standard regression techniques using this data failed to produce reliable models for UVI prediction. The imprecision in some of the variables and small sample size implied that much more sophisticated techniques should be used. In the current research we resorted to artificial neural networks (ANNs) to cluster the data and then to model the UVI estimate in each of the data clusters. In the ANN training, we utilized the weights pruning method of optimal brain surgeon type to enable good generalization of the ANN models. The results obtained in the UVI assessment by this method produced results of significant accuracy.