Statistical analysis of massive datasets very often implies expensive linear\nalgebra operations with large dense matrices. Typical tasks are an estimation\nof unknown parameters of the underlying statistical model and prediction of\nmissing values. We developed the H-MLE procedure, which solves these tasks. The\nunknown parameters can be estimated by maximizing the joint Gaussian\nlog-likelihood function, which depends on a covariance matrix. To decrease high\ncomputational cost, we approximate the covariance matrix in the hierarchical\n(H-) matrix format. The H-matrix technique allows us to work with inhomogeneous\ncovariance matrices and almost arbitrary locations. Especially, H-matrices can\nbe applied in cases when the matrices under consideration are dense and\nunstructured.\n For validation purposes, we implemented three machine learning methods: the\nk-nearest neighbors (kNN), random forest, and deep neural network. The best\nresults (for the given datasets) were obtained by the kNN method with three or\nseven neighbors depending on the dataset. The results computed with the H-MLE\nmethod were compared with the results obtained by the kNN method.\n The developed H-matrix code and all datasets are freely available online.\n