the infiltration of hazardous chemicals into the soil causes soil pollution which poses significant risks to ecosystems and human health. For this reason, accurate predicting the diffusion of pollution in soils is important and critical for monitoring and protection the environmental state. In this study we have compared advanced machine learning ML models to predict vertical and horizontal pollution diffusion using complex and multimodal soil experimental datasets. Support vector regression, linear regression, gradient boosting regression, xgboost regression, k-nearest neighbours, and artificial neural networks were employed to build predicted models and compared with each other. The comparison criteria are measuring mean squared error, root mean squared error, mean absolute error, and R-squared as the metrics used to evaluate the predictive models performance. The observed results demonstrate that ensemble methods XGBoost and random forest, outperform other models in predicting pollution diffusion while XGBoost achieving the highest accuracy. On the other hand, linear regression was the least effective while k-nearest neighbours and artificial neural networks showed moderate performance.