Abstract
Computational methods were used to reduce the dimensionality and to find clusters of multivariate data. The variables were the natural radioactivity contents and the texture characteristics of sand samples. The application of discriminate analysis revealed that samples with high negative values of the former score have the highest contamination with black sand. Principal component analysis (PCA) revealed that radioactivity concentrations alone are sufficient for the classification. Rough set analysis (RSA) showed that the concentration of 238U, 226Ra or 232Th, combined with the concentration of 40K, can specify the clusters and characteristics of the sand. Both PCA and RSA show that 238U, 226Ra and 232Th behave similarly. RSA revealed that one or two of them can be omitted without degrading predictions.