The number of objects in space is increasing over time, and therefore it is desired to find more efficient propagation models in terms of accuracy and computational speed. This thesis project focuses on the propagation of the state and uncertainties of Potential Hazardous Asteroids, to predict potential Earth impacts. A machine learning technique is proposed to reduce the computational expensiveness of current propagation techniques. The PHA position and the error made by the uncertainties are predicted using a neural network, which uses data from currently known PHAs. Several options are considered regarding the input and output variables and the networks are also tuned. The aim for the best model is to find a reasonable accuracy for the prediction of the position and the uncertainty of newly discovered PHAs.