We focus on the analysis of human mobility data for planning public and private transportation systems. In both cases, we make use of massive real life GPS datasets, obtained from subscribers of a pay-as-you-drive car insurance contract, under which the vehicle trajectories are periodically sent (through the GSM network) to a central server for anti-fraud and anti-theft purposes. This dataset has been donated to KDDLAB - UNIPI for research purposes by a leading company in Europe. The dataset contains about 40,000 cars tracked during 5 weeks (from June 14th through July 18th, 2010) in coastal Tuscany, a 100 km x 100 km square centered on the city of Pisa. Other additional data about the public transportation systems concern their schedules, their routinely collected real usage logs and/or customer surveys.

Public Transportation Planning: Our first goal is to plan the schedules, the costs, and the connections of public transport systems (bus, taxi, trains, shared cars and bikes), finding out a good trade-off between the total cost of the proposed solution and the quality of the service offered. In this scenario, it is clear the presence of a number of hard constraints, which include the street network (structure, speed limits, etc.), the impact on the environment (maximal amount of pollutants produced as required by law, etc.), the minimum granted level of service, and so on, together with soft constraints such as matching the actual mobility needs, lowering costs (for both the providers and the customers), etc. Constraint programming is used in order to produce a first version of a schedule for the public transportation systems. Driven by background knowledge expressed as spatio-temporal and frequency constraints, we subsequently search meaningful patterns of movements performed by private vehicles.  By comparing the two kinds of patterns (private vs public), we derive knowledge about the gap between the needs emerging from data and the actual service provided. This knowledge shall then be translated into a set of constraints representing the difference between the two models, which is then the basis for revising the set of constraints on the public transportation planning.

Car Pooling: The second goal is to plan schedules, on a per-vehicle basis, for sharing of private cars over trips among two or more persons. For each vehicle, we first extract mobility profiles of its frequently-followed trajectories by data mining. Matchings of profiles and possible schedules are then determined by resorting to constraint satisfaction and optimization. These intertwined steps are iterated by simulating the effects on data of the adoption of (some of) the suggested schedules. Moreover, we will also consider the network effects involved in the process: how similar people tend to follow similar movement behaviours, what are the social factors influencing the results of the car pooling schedule, how mobility influences social connections. The patterns of of network effects are, again, used to revise and improve suggested schedules.