In this work, a bi-directional LSTM has been employed to predict the future positions of a maritime vessel, known as the target, based on noisy estimates of its previous and current positions. In a first set of experiments, plain trajectories generated by a simulation were used for training and noisy trajectories were used for testing. In a second set of experiments, noisy trajectories were used for training and testing. The accuracy and the loss achieved in both sets of experiments have demonstrated that this type of network is capable of solving the Target Motion Analysis problem.