Purpose <p>Soil organic matter (SOM) is an important soil quality parameter and plays a central role in soil nutrient cycling, aggregate stability and water retention. This study aims to investigate the effectiveness of different machine learning algorithms in the simulation of SOM in agricultural soils and to explore the role of environmental factors in SOM simulation.</p> Materials and methods <p>In this study, 223 soil sampling points were obtained, and the topography, the Normalized Difference Vegetation Index (NDVI) and SOM information at adjacent points were selected as environmental factors. The spatial simulation and prediction of SOM were carried out using the particle swarm optimization-back propagation neural network (PSO-BPNN), particle swarm optimization-random forest (PSO-RF), particle swarm optimization-support vector machine (PSO-SVM), backpropagation neural network (BPNN), random forest (RF), and support vector machine (SVM) methods.</p> Results and discussion <p>The validity of the models was verified by MAE, MRE, RMSE, MAPE and R², and the prediction accuracies of the six models were PSO-SVM &gt; PSO-BPNN &gt; PSO-RF &gt; RF &gt; SVM &gt; BPNN in descending order. The particle swarm algorithm significantly improved the computational accuracies of the SVM, BPNN and RF models, in which the R² values increased by 16.36%, 18.87% and 3.28%, respectively. The PSO algorithm was found to play an optimization role in the BPNN, RF and SVM methods, speeding up the process of obtaining the optimal solution and improving the speed of iterative convergence and the prediction efficiency.</p> Conclusions <p>The combination of the particle swarm algorithm and SVM, BPNN and RF modeling achieved good results in the spatial prediction of SOM in the cultivated land of the plain area. As a result, our findings aid in optimizing soil sample density and serve as a foundation for future SOM digital soil mapping approaches (DSM) in comparable regions.</p>