The purpose of this research is to model the water permeability of soil at different levels of mud contamination using learning algorithms. In this research, 3 texture classes of clay, silt and sand were used at concentrations of 2, 5 and 10 g/l to create mud contamination and similar to flood flows in 4 repetitions. Also in this research, the physical properties of the soil of the region including soil texture, porosity, matric potential, soil bulk density and soil mass and volumetric moisture were determined. Then, using the measured data, the cumulative infiltration values, maximum infiltration rate, average infiltration rate, final infiltration rate, time to infiltration stabilization moment and infiltration reduction slope were calculated and then for permeability modeling using learning algorithms and determining the best model, MLP model and SVM and GP models with PUK and RBF kernels were used. The models were evaluated using evaluation criteria including root mean square error, coefficient of determination and correlation coefficient. The results of comparing the infiltration rate in different soil textures with different concentrations showed that the infiltration rate was more affected by sand-sized suspended solids compared to clay-sized and silt-sized suspended solids. The modeling results in the different treatments studied showed relatively good performance of all models and it was found that the SVM and GP models with PUK kernel performed better than the RBF kernel. According to these results, in the control treatments, 2 grams of clay and 2 grams of silt, the SVM-PUK model and in the treatments of 5 grams of clay and 2 grams of sand, the GP-PUK model had higher performance than other models. These models with high coefficient of determination and correlation coefficient as well as low root mean square error were selected as superior models compared to other models in this study