Principal component analysis (PCA) is a solution to the multicollinearity problem of dam regression models. However

uninformative principal components (PCs) may lead to the failure prediction of dam deformation. Thus a hybrid regression model using semi-parametric regression and PCA is proposed

where the PCs with the highest variance are treated as the semi-parametric component; the remaining PCs and model errors are treated as the non-parametric component to be estimated. The hybrid model is tested using the field observations of a dam in China. The result shows the hybrid model can circumvent the multicollinearity of dam causative effects and the ill-conditioned problem in semi-parametric penalized least squares regression. A comparative study with traditional PC regression and stepwise regression demonstrate the superior performance for dam deformation prediction