This work focuses on a smart mobility use case where real-time data analytics on traffic measures is used to improve mobility in the event of a perturbation causing congestion in a local urban area. The data monitored is analysed in order to identify patterns that are used to properly reconfigure traffic lights. The monitoring and data analytics infrastructure is based on a hierarchical distributed architecture that allows placing data analytics processes such as machine learning close to the source of data. This distributed data analytics architecture is presented and specified for the targeted use case. Aiming at detecting traffic perturbations, several classifiers are proposed, implemented and trained with stationary traffic data. An optimization method for traffic light phase reconfiguration is proposed. The accuracy of classifiers and traffic optimization are then evaluated in a non-stationary environment where perturbations gradually arise from traffic normality. To this purpose, the open-source traffic simulator SUMO is used.