Multicollinearity and difficulty of interpreting the coefficients of dam regression models pose two problems: (1) selection of informative variables for analysing dam deformation behaviour, and (2) mitigation of the multicollinearity among the variables. Resolving these two problems necessitates the application of genetic algorithm-based partial least square (GA-PLS) and statistically inspired modification of PLS algorithm (SIMPLS). A SIMPLS regression with the predictor variables selected by GA-PLS (hybrid GA/SIMPLS regression) is put forward to interpret the results obtained from periodic monitoring surveys of hydraulic structures. The hybrid model is employed for analysing the crack behaviour of an earth-rock dam in China. The results show the proposed model is superior to an ordinary SIMPLS and stepwise regression, especially when multicollinearity and influential outliers exist among the variables.