The goal of this thesis was to train a classification and several regression models to solve a number of problems. A large amount of data in form of simulations was collected on BURA supercomputer that was used to train the models. The classification models were trained with the purpose to precisely localize the occuring leak while the regression models were trained to predict the value of the area, discharge coefficient and strating time of the leak. In the context of classification we obtained very similar results between two algorithms so the results for both of them are presented with attention to detail. The results obtained by the regression models via cross validation are presented in terms of R2 coefficient and value of the mean squared error with regards to the number of simulations used for training. The data is split in to three sets that were obtained by runinng a very large number of simulations on the EPANET Net3 network and on the Network 1 from [28]. Two sets of data were collected for Network 1 with two different sensor placements.