Mining user behaviours from taxi rides and location based social networks using machine learning techniques

Nowadays, the large availability of smart city data allows the analysis of users’ preferences and behaviors inside a town. The opportunity of providing a framework about the frequency of user check-ins in some places of a city might have a key role in maintenance, cleaning, security, and economic growth. This thesis work aims at exploiting a machine learning based approach to extract useful patterns from mobility data and location-based social network data. The proposed methodology addresses two main issues. The first one is related to suggest locations that users may appreciate, while the second regards an estimation of users’ displacements. The first problem implies the adoption of a clustering algorithm and a matrix factorization algorithm. This strategy involves the exploitation of some features, such as the geo-coordinates of the user location, the category of the associated venue, and the hour and the day of the week in which a given location in the city has been explored by the user. Based on these features, it is possible to split users into disjoint groups and retrieve useful insights. For example, a large number of users result to be tourists, but there is also a not negligible portion of commuter workers. The second problem has been addressed by applying the C-Spade algorithm to extract sequence patterns from taxi trips data. A preprocessing step has been adopted to first organize taxi trips data, and then outline a grid of the points of interest located within a given radius from a taxi path. The work presented in this thesis can be of interests in many different contexts. For example, for tourism forecasting, but also for public transport operators, since the acquired knowledge might leadto optimize transport in the most frequented areas. In fact, transportation, in large cities, remains one of the fields which still have to cope with different challenges.