Electronic Thesis and Dissertation Repository


Master of Engineering Science


Electrical and Computer Engineering


Miriam A.M. Capretz


The availability of location data increases every day and brings the opportunity to mine these data and extract valuable knowledge about human behaviour. More specifically, these data may contain information about users’ activities, which can enable, for example, services to improve advertising campaigns or enhance the user experience of a mobile application. However, several techniques ignore the fact that users’ context other than location and time, such as weather conditions, influences their behaviour. Moreover, several studies focus only on a single data source, addressing either data collected without any type of user interaction, such as GPS data, or data spontaneously shared by the user, for instance, from location-based social networks (LBSNs), but not both.

This thesis proposes a framework that aims to predict users’ current activity preferences (UCAP). UCAP handles data gathered from different sources. It takes into account users’ historical data, their current context, and other external contexts such as weather conditions.

The framework was evaluated on five real-world datasets. The results demonstrated the accuracy of the proposed solution, which was on average 12.3% more accurate than a state-of-the-art technique. Moreover, the experiments evaluated the impact of the main components on the prediction results and showed that UCAP is not constrained by dataset size.