The paper presents a microscopic modeling of surface public transport travel time. Results are performed on data collected with the DIALEXIS tool, which enables very precise measurement of the vehicle travel time at each step. We proposed hierarchical modeling; firstly, machine-learning techniques are used to find the most influencing set of components among the waiting time while doors are closed. Then to the global travel time, including the waiting time while doors are open and the running time. We compared the results of the LASSO and the Random Forest Regression methods. After retrieving the results and evaluating the models, we applied the graph algorithm of PageRank, then we trained the generated importance coefficients. Finally, we evaluated and compared all the models on two datasets, a Rapid Bus Transit and a normal bus. Further to the travel time modeling the paper shows that graphs can be used to feed machine learning models and find new features to use for training, subsequently speeding up artificial intelligence decisions. We also concluded that the Random Forest model is most performant and robust than the LASSO.