Absolute permeability is an important transport property of a porous medium that requires determination on special equipment, so its determination is an important task. This article examines the effectiveness of machine learning regression algorithms for predicting the absolute permeability of various rocks. The performance of algorithms such as Random Forest, Gradient Boosting, Support Vector, Lasso, k-Nearest Neighbors, and Gaussian Process was compared based on data set of 266 sub-samples of carbonate and sand rocks, as well as artificial sand packing. Properties of each sub-sample such as pore radius, pore throat radius, coordination number, porosity, specific surface area, tortuosity and absolute permeability are extracted using pore-network simulation of the single-phase fluid flow through each sub-sample. The influence of the training/testing data subset (70/30 and 80/20) and the number of features of input data set on the performance of each of the above algorithms was investigated. The results showed that for the considered data set, the Random Forest algorithm was the most suitable for predicting absolute permeability with high accuracy. The highest predictive accuracy was R²=0.83, and it was obtained using 5 out of 6 features of input dataset. The Gradient Boosting algorithm also showed good predictive ability for absolute permeability, although it chose almost one feature (porosity) as important. Its highest accuracy was R²=0.73 at 80/20. The results of the study also showed that all algorithms except Random Forest predicted significantly higher minimum permeabilities. Also, all algorithms, except of Support Vector and k-Nearest Neighbors, predicted the mean permeability with the minmal errors.