Hyper-parameters optimization and validation of brain image regression

Based on the non-homogeneity validation method, the hyper-parameter optimization of general regress model on fMRI brain data is analyzed. To acknowledge the abilities of different regression methods we compare ridge regression, support vector regression and Elman recurrent neural network over several feature rating sequences. The original data is obtained from PBAIC2006. Experiment results show that this method has good stability and generalization; also it can be used in the field which lack of knowledge of the relevant fields. Also, we find the linear method is simple, effective on fMRI regress missions, and higher predictive ability but lower calculating costs than other methods on many feature rating sequences.