To achieve rapid and accurate determination of the optimal compaction frequency for high-speed railway subgrade materials, a method based on the PSO-BPNN-AdaBoost model for intelligent frequency estimation is proposed. Firstly, the Particle Swarm Optimization (PSO) algorithm is introduced to obtain the optimal hyperparameters of the Backpropagation Neural Network (BPNN), and then the PSO-BPNN-AdaBoost model is established by integrating the AdaBoost ensemble algorithm. Secondly, taking graded gravel fill material as an example, the Grey Relational Analysis algorithm (GRA) is employed to identify the main controlling features affecting f op as input features for the model, and the predictive performance of the model is evaluated. Finally, the model’s reliability is verified through ablation analysis. The results indicate that the PSO-BPNN-AdaBoost model demonstrates higher predictive accuracy. The main controlling features influencing f op are revealed to be the maximum particle size (d max), gradation parameters (b, m), coarse aggregate elongation index (EI), Los Angeles Abrasion (LAA), water absorption rates (W ac, W af).