Hydromica is a typical alteration mineral in granite-type uranium deposit, and also an important indication of uranium. The amount of hydromica to some extent reflects the strength of hydromicasization in uranium deposit. Because of the bad performance of the traditional modelling methods in prediction, in the present paper, the authors’ adopt SMOreg in the spectral modelling for hydromica, and validate its effectiveness. The authors’ also propose a novel method called ICSMOreg. In this method the authors’ employ instance cloned method to learn the samples selected by having a strong affinity with the test sets, and then get the new samples into SMOreg to build the spectral model. Finally, we experimentally compare ICSMOreg with SMOreg, artificial neural network, model tree and the common modelling methods like linear regression, multiple linear regression. The result shows that the new method improves the accuracy of prediction, and also reduces the negative impact of noise.