Urban mobility management has gained significance recently in order to help cities achieve their sustainability goals and improve traffic efficiency. For those purposes, authorities and other institutions require decision-making tools, and these tools are based substantially on traffic data. In other words, if the typology of traffic is known, or even predicted, actions can be taken to minimize its negative effects. In fact, the large volume of data available today is useful to anticipate situations and overcome mobility issues. In this paper, vehicle volume measured by induction loops of the city of Santander (Spain) on urban roads is analysed and predicted, comparing results between different models, input sequences and horizons. Firstly, data is imputed to supply possible lack of data, and other variables such as holidays or day of the week are added to enhance the performance. Secondly, data is analysed with a descriptive statistical study. Lastly, deep learning models applied to time series are trained and predictions of vehicle flow are inferred, based on historical data and trying different combinations of input and output sequences. Some of these models achieve high accuracy (with best results reaching a 10% of MAPE in hourly vehicle flow), allowing transportation managers understand the mobility of their cities and helping them take informed decisions about future events based on these insights. This high accuracy, indeed, allows forecasting traffic reliably, so that measures can be safely chosen.