The linear relationship among variables characterizing sunspot group characteristics and their flare activity was commonly used for construction of the prediction functions. Hirman et al. (1980) and Neidig et al. (1986) included as input parameters also such variables which were combinations of certain choosen characteristics. Jakimiec and Bartkowiak determined prediction functions based on the quadratic relationships and obtained higher values of determination coefficient (used as an index of the prediction quality) than those for linear relationships. More comprehensive investigation of the interrelation structure of quadratic relationships among original characteristics was not
performed as yet. In this paper we apply correlation matrix analysis and cluster analysis method for study of the interrelation structure among p=13 variables characterizing D E F sunspot groups observed in 1979 year. Basing on the quadratic relationships of the variables (for p=13 we have to include into calculations p'=104 variables) and applying regression methods we obtained the predicted functions. Total X-ray flare fluxes for 1-8 k and for 0.5-4 k are used as thepredicted characteristics of flare activity. We found that the quadratic relationships provided higher determination coefficients than the linear relationships do. That means that the products of input characteristics provide closer correlation with the flare activity the next day than the input characteristics alone. However, when the predicted functions are extrapolated, i.e. when the real forecastions are performed for another data set (for sunspot groups observed in the first half of 1980 year), the prediction quality is much better for the linear prediction functions.
This result indicates that the structure of quadratic relationships among sunspot group characteristics is less stable than the structure of linear relationships is. That means for real predictions other sophisticated models should be used. We discuss one such models.