BOOSTING PREDICTIONS OF FREE-FLOATING BIKE SHARING SYSTEMS WITH SPATIAL CLUSTERING TECHNIQUES

There exist various types of public transports in the cities, starting from trains, buses, trams to the newest rental bikes and electric scooters. The use of the latter is even more helpful to the ecosystem, given that no carbon dioxide is produced during use and they have a lot of potential to improve the city's environment, making it eco-sustainable and smart. The stationless bike sharing system, called floating, is replacing the one that uses fixed stations, called docking, for a matter of ease of use and freedom left to the end user. A bike can be left in any place you choose, even a few meters from the destination, without having to look for a mandatory delivery area. One of the remaining problems is the availability of means of transportation in case of need. These are distributed through the use of vans that collect and place them in defined areas, so that they can be found available and sorted as needed. The aim of this thesis is to identify areas of use of this type of bikes through various spatial clustering techniques, using the city of Padua (Italy) as center of the study. In addition, a rectangle division of the analysis area was used to go deeper and further highlight differences. Subsequently, two prediction methodologies were implemented to identify the number of vehicles that should be made available in a specific region and in a specific time slot. ARIMA and XGBoost were used as prediction technologies, which allow in a reliable and precise way to have a real value of means of transportation that will be used in that time period. The various predictions were then compared using statistical methods to analyze the best clustering method and the number of ideal regions to create. The application of this method of analysis, study and prediction can be used to make the distribution of bicycles more optimized in order to make this type of transport more sustainable for the environment.