Towards User Behavior Forecasting in Mobile Crowdsensing Applications.

Mobile crowdsensing has rapidly become an interesting and useful methodology to collect data in modern smart cities, thanks to the pervasiveness of users mobile devices. Although there are many different proposals, opportunistic and participatory mobile crowdsensing are the most popular ones. They share a common goal, but require a different effort from the user, which often results in increased costs for the service provider. In this work we forecast user participation in mobile crowdsensing by leveraging a large dataset obtained from a real world application, which is key to understand whether there are areas in a city which need additional data obtained through raised incentives for participants or by other means. We then build a custom regressor trained on the dataset we have, which spans across several years in different cities in Italy, to predict the amount of reports in a given area at a given time. This allows service providers to preventively issue participatory tasks for workers in areas which do not meet a minimum number of measurements. Our results indicate that our model is able to predict the number of reports in an area with an average mean error depending on the precision needed, in the order of 10% for areas with a low number of reports.