Abstract: In order to support dam safety evaluation and optimization design,deformation behaviors are usually required to be quickly estimated in the design and construction process of the concrete face rockfill dam (CFRD).Deformation evaluation and control are key issues in the CFRDs construction.This paper combines threshold regression (TR) and improved support vector machine (SVM) algorithm to establish intelligent prediction models for CFRDs typical deformation behaviors.Firstly,the measured data of 87 CFRDs are collected.Based on the statistical review of the typical deformation behavior,the mathematical relationship between three typical deformation indexes and six influencing factors is established adopting multiple linear regression theory.The main influence factors are analyzed.Considering the characteristics of non-linear mutation and discreteness of the case data,multivariate TR theory is used to cluster the deformation data according to the dam height.A hybrid weight coefficient is introduced to construct an adaptive hybrid kernel function and the particle swarm optimization algorithm is used to determine main parameters of the SVM.Then the improved SVM prediction model is established in the clustering interval of different dam heights.The comparative analysis of the prediction results of the established model and the existing prediction models shows that the model has good prediction accuracy and can be used to accurately predict CFRD typical deformation behaviors.