Abstract

Every day, billions of mobile network log data (commonly defined as Call Detailed Records, or CDRs) are generated by cell phones operators. These data provide inspiring insights about human actions and behaviors, which are essentials in the development of context aware appli- cations and services. This potential demand has fostered major research activities in a variety of domains such as social and economic development, urban planning, and health prevention. The major challenge of this thesis is to interpret CDR for human activity recognition, in the light of background knowledge of the CDR data context. Indeed each entry of the CDR is as- sociated with a context, which describes the temporal and spatial location of the user when a particular network data has been generated by his/her mobile devices. Knowing, by leveraging available Web 2.0 data sources, (e.g., Openstreetmap) this research thesis proposes to develop a novel model from combination of logical and statistical reasoning standpoints for enabling human activity inference in qualitative terms. The results aimed at compiling human behavior predictions into sets of classification tasks in the CDRs. Our research results show that Point of Interest (POI)s are a good proxy for predicting the content of human activities in an area. So the model is proven to be effective for predicting the context of human activity, when its total level could be efficiently observed from cell phone data records.