Demographic Attributes Prediction on the Real-World Mobile Data

The deluge of the data generated by mobile phone devices imposes new challenges on the data mining community. User activities recorded by mobile phones could be useful for uncovering behavioral patterns. An interesting question is whether patterns in mobile phone usage can reveal demographic characteristics of the user? Demographic information about gender, age, marital status, job type, etc. is a key for applications with customer centric strategies. In this paper, we describe our approach to feature extraction from raw data and building predictive models for the task of demographic attributes predictions. We experimented with graph based representation of users inferred from similarity of their feature vectors, feature selections and classifications algorithms. Our work contributes to the Nokia Mobile Data Challenge (MDC) in the endeavor of exploring the real-world mobile data.