Abstract: In the mineral resources quantitative prediction using the least squares support vector machine (LS-SVM)， precision of results are influenced by the selection of its parameters. The prediction method based on the LS-SVM combining with Bayesian inference is proposed and it is also compared with weights-of-evidence (WofE) method. During the training process， the optimized parameters of LS-SVM are chosen by Bayesian inference method， which can build the optimized model for the mineral resources quantitative prediction. The results show that the proposed method not only overcomes randomness and limitation of its optimal parameter selection， but also increases the accuracy of prediction by exporting the prediction result in the form of posterior probability.

Key words: Bayesian inference, LS-SVM, mineral resource, quantitative prediction, WofE(weights-of-evidence)