This dissertation is situated within the field of Mathematical Modeling, with an emphasis on problems related to Reservoir Engineering. In particular, the text explores the application of neural networks for the prediction of key oil well data, such as bottom-hole pressure, oil flow rate, and water flow rate over extended periods. To achieve this, the method known as Embed to Control and Observe was employed. One of the main topics discussed in the study concerns the sensitivity of neural network hyperparameters, which are defined during the training process. Specifically, the investigation focused on how variations in these hyperparameters affect the accuracy of the predictions. It was observed that the weights assigned to the cost functions (transition, output transition, water flow rate, saturation in producing gridblocks), the batch size, the seed, and the Python version significantly influenced the prediction accuracy.