In the present study, intelligent methods including artificial neural network (ANN) support vector machine (SVM) optimization and their combinations artificial neural network-particle swarm optimization (ANN-PSO), wavelet-artificial neural network (W-ANN), and W-ANN-PSO were investigated to predict the performance of rockfill dam crest settlements. Input parameters were based on the crest settlement data from a rockfill dam with a central core and the dam height and compressibility index. The results showed that the artificial neural network with 66% accuracy is the basis of the effectiveness of the optimization process and data preprocessing. The minimum error values by the neural network method are 1.88%, and the maximum value is 37.44%. Also, the average error was 14.23%. SVM optimization method and radial basis function (RBF) performance are often superior to other functions due to their radial nature. The reason for the greater compatibility of RBF performance and better fit to data is the lower absolute mean error value compared to other methods. With the ANN-PSO method, the maximum error is 11.2%, the minimum error value is 1.17%, and the average is 4.66%. By examining the validation data, it can be concluded that the errors are consistently in the range of 5–11%. The preprocessed neural network method and the performance of the bior 6.8 wavelet function has superior performance compared to other W-ANN models, so its average absolute error is about 29%. The db4 wavelet function performs better than other functions in the W-ANN-PSO model. The W-ANN-PSO model performed better than the model without PSO optimizer because the particle aggregation method dealt with complexity by increasing the number of inputs to the neural network and reducing their effects.
