Driver behavior is the set of actions that a road user undertakes during a driving task. There is a huge interest in studying driver behaviors evaluating fuel consumption and improving safety. Using in-vehicle sensors is a widely adopted methodology to fulfill these objectives. The large amount of data generated from multiple sensors opens the doors to machine learning and deep learning algorithms which are more adequate than other methodologies. In this paper, a Feed Forward Neural Network is trained and tested with OBD and geographic data to classify driver behaviors. Several datasets are analyzed but the most adequate has resulted in the DDD20 dataset. This contains a larger amount of data than the other one, with 51 h and 4000 km of total driving times and distances. After the selection of the dataset and the enrichment with geographical data, feature selection, and data labeling techniques and algorithms are implemented. The model shows a high level of accuracy (above 98%) for the three classes of driver behavior studied.