This paper proposes a real time travel time prediction framework designed for large urban area including both arterial and urban roads. This framework makes it possible to test a wide variety of prediction models based either on theoretical or data-driven approaches. The results are demonstrated in a large test case corresponding to the Amsterdam Practical Trial. Data-driven approaches were then favor because their are easier to calibrate and require less computations. For short-term prediction, it appears that the simplest data driven approach (naive approach) performs the best. For larger-time window, a refined method (historic median prediction) provides the more accurate results. In most cases, the average absolute relative error is below 20%. The main contributions of this paper are (i) the formulation of the global framework and (ii) the extensive test of different methods on a large and heterogeneous operational test cases. The operational feedbacks from this study provide a good state of the art of the performance of data-driven methods in a mixed context and pave the way of further methodological developments