Table 2.

Quality measures (accuracy and κ coefficient) of multi-layer perceptron MLP, radial basis function network RBF, k nearest neighbors k-NN and ridge regression RR

	MLP	RBF	k-NN	RR
accuracy [%]	97,35	88,41	96,58	94,92
κ coefficient	0,966	0,845	0,956	0,934

These were evaluated using test set (DatasetF_test). The MLP model uses geometric preprocessing with the order k = 1, has 22 hidden neurons with the log-sigmoid transfer function and output neurons use the tan-sigmoid transfer function. The best RBF model uses geometric preprocessing with the order k = 1, has 18 hidden neurons with the spread of 0.15. The optimal value of k in k-NN is 11. In RR, the optimal regularization parameter λ is zero, and the order of the geometric preprocessing expansion k is 6.