Human activity recognition from spatial data sources

Recent availability of big data of digital traces of human activity boosted research on human behavior. However, in most of the datasets such as mobile phone data or GPS traces, an important layer of information is typically missing: providing an extensive information of when and where people go typically does not allow understanding of what they do there. Predicting the context of human behavior in such cases where such information is not directly available from the data is a complex task that addresses context recognition problems. To fill in the contextual information for such data, we developed an ontological and stochastic model (HRBModel) that interprets semantic (high-level) human behaviors from geographical maps like OpenStreetMap, analyzing the distribution of Points of Interest (POIs), in a given region and time period. The semantic human behaviors are human activities that are accompanied by their likelihood, depending on their location and time. In this paper, we perform an extended evaluation of this model based on other qualitative data source, namely a country-wide anonymized bank card transaction data in Spain, which contains contextual information about the locations and the types of business categories where transactions occurred. This allows us to validate the model, by matching our predicted activity patterns with the actually observed ones, so that it can be later applied to the cases where such information is unavailable. This extended evaluation aimed to define the applicability of the predictive model, HRBModel, taking various type of spatial and temporal factors into account.