The annual cost estimate of traffic congestion exceeded 88 billion dollars in 2019 in the United States. While it consistently alerts alleviating congestion, the importance of comprehending travel behavior arises concurrently. Demand correlates with individual travel decisions and is directly associated with network performance. In that, travel demand is a congregated product from multi-dimensional individual choices. Owing to the advanced granularity of human mobility data and computational capability in recent years, a real-time, personalized decision support system has become a realistic alternative to improving the quality of road network performance and individual travel experience. The emergence of mobile device location data (MDLD) has opened new research avenues that provide in-depth information on individual travel behavior. This data-driven evidence is vital for travel demand analysis related to road network structure because it helps to 1) directly measure genuine travel decisions conducted by individuals; 2) consider heterogeneous behavioral preference among travelers; and 3) capture interconnected relationships between travelers’ perspectives and spatiotemporal characteristics within the transportation system, all of which were scarcely feasible without individual mobility evidence. The research effort in this dissertation seeks to best leverage these high granularities of data with machine learning algorithms. A data-driven framework for modeling route choice behavior is proposed, capable of learning historical trips and gradually evolving to future route choice decisions. The dissertation also explores mobility trends from a sizeable MDLD sample set. It provides a diverse spectrum of empirically supported proof to relax the rigid framework of traditional demand analysis. The contribution of the dissertation is four-fold: 1) to provide empirical support on route choice behavior in a dense urban network; 2) to introduce an alternative set generation method that measures semantic road importance and reflects individual choice patterns; 3) to demonstrate a data-driven generalized and personalized route choice prediction model that considers individual preference; and 4) to propose a treatment method for low-data issues within the person-specific model, named Doppelgänger Finder. Broadly speaking, the proposed framework is expected to contribute to the forthcoming modeling evolution—from an aggregated and simulated to a personalized and observed scheme—and facilitate person-specific decision support.