Least Squares Support Vector Regression (LSSVR) machine has the advantages of small sample, nonlinearity and high dimensionality, which can solve the problem of predicting the compressive strength of green concrete with oil palm shell aggregate. In this paper, the error sum of squares instead of the error sum is used as the objective function, IFFA is used to find the optimization of the kernel function parameters and penalty factors of LSSVR, and the PWLCM-based chaotic search is used to initialize the population, and ultimately the improved auricular fox algorithm is realized for the optimization of the least squares support vector regression algorithm, which makes it have strong fitting and generalization abilities, and significantly reduces the burden of computation, thus improving the Computational efficiency. Application of the designed combined algorithm for compressive strength prediction of concrete reveals that the R², MAPE and RMSE values obtained by this paper’s model on the training dataset are 98.71%, 5.92% and 1.0823 MPa, respectively. The correlation coefficients predicted by the model are much closer to 1 as compared to that of the baseline model, which suggests that this paper’s model possesses a superior generalization capability, making it more effective in dealing with complex and invisible data. The adopted method is practical and innovative, and is of guiding significance for practical engineering.