In this project we are developing orbital optimizations for the SCF and DFT procedures. Efficient and robust orbital optimization procedures are mandatory in computational chemistry. In this project we are exploring improvements to three different methods of orbital optimization. First, the so-called direct inverse iterative subspace method (DIIS) is considered. Second, the restricted-step rational function optimization procedure is used. Finally, a novel implementation of a subspace gradient enhanced kriging (a Gaussian Process Regression-like approach -- a machine learning method adapted to the SCF/DFT procedure) is for the first time explored. The development require benchmark calculation on a large number of molecular systems to demonstrate the characteristics and performance of each of the methods. In this project we use 2080 different molecules which are representatives of organic molecules and transition metal complexes in various spin states.