The goal of this thesis is to analyze lane changes and lane choice in a congested urban envi- ronment. This is possible thanks to the pNEUMA dataset, a new trac dataset resulting from a eld experiment with 10 drones ying over the central district of Athens, Greece. Using an already available lane detection algorithm the trac information at the lane level is extracted. The study on lane-changing activity reveals the con icting potential of lane merges and the particularity of motorcycle behavior. Motorcycles perform more lane changes on average than other vehicles and evidence on the ltering phenomenon is provided. On the other hand, the analysis on driver lane choice shows the dierent behavior of left- and right-turning vehicles, the latter being in uenced by the friction elements of the right edge of the street. Finally, the last part of the thesis uses machine learning classiers to predict the driver's intention to turn using the previous analysis on lane changing and lane choice. iii