This papers presents the design of the travel mode detection component within a generic architecture of processing individual mobility data. It approaches mode detection in two steps, each aiming at a particular objective. The first step develops a discriminative classifier that detects the mode of the observed trips or a sequence of modes in a multiple leg journey. It requires a considerable amount of ground truth data with known modes to be available for training. It also relies on a k-shortest path algorithm that generates plausible alternatives routes for the journey. The second step utilizes the discriminative recognition step of the observed mode in order to build a behaviorally grounded model that predicts the chosen mode within a set of available alternatives as a function of user characteristics and transportation system variables. It is based on the discrete choice modelling paradigm and results in a set of parameters calibrated for distinct neighborhoods and/or segments of population. The overall framework therefore enables travel mode choice modeling and a consequent policy analysis and transportation planning scenario evaluation by leveraging privacy-sensitive individual mobility data possibly held in a secure private repository. It provides a set of algorithms to drastically reduce the latency and costs of obtaining a crucial information for models used in transportation planning practices. The performance and accuracy of the algorithms is evaluated experimentally within a large metropolitan region of the San Francisco Bay Area.