Predicting an aircraft's Estimated Time of Arrival (ETA) while enroute can be a challenging endeavor. The great number of factors that can affect a flight's punctuality range from things well under the pilot's control, such as flight level and cruise airspeed, all the way to environmental circumstances that are generally very hard to predict, such as weather phenomena and airport congestion. Therefore, aircraft ETA predictions tend to rely heavily on aircraft performance models, along with either parametric or physics-based trajectory models, being only sometimes enhanced by simplistic statistical considerations, such as the average winds encountered in a flight path during a certain period of the year. This work presents a method for enhancing aircraft ETA predictions by applying machine learning techniques, taking into account general information about the flight as well as weather and air traffic. A good amount of effort is put into feature generation and selection, and subsequently a model is built from representative flight, weather and air traffic data, allowing for an increase in prediction accuracy. Some of the challenges that arise from the nature of the data are discussed, such as the fact that weather information is naturally fragmented into a great number of variables, which makes it difficult to extract value from it without a very large number of samples covering all possible scenarios. The results show that it is possible to enhance the ETA predictions obtained from traditional methods by correcting them with a model that takes into account the statistical relationships observed between flight, air traffic and weather information.