Travel behavior lies at the core of analysis and evaluation of transportation related measures aiming to improve urban mobility, environmental quality and a wide variety of social objectives. A better understanding of travel behavior will improve travel demand forecasting and the assessment of emerging transport policies, and will improve our means to increase road safety. The day-to-day models reflect the travelers' learning and forecasting mechanisms. These models predict travelers' choices for any given day based on their experienced choices in the previous days. Day-to-day approaches allow the use of wide range of behavioral rules, and levels of aggregation, and capture the heterogeneity in users' learning and adaptation processes, and behavioral characteristics. This thesis aims to develop a novel framework to model the interdependence between travelers' choice decisions, learning and adaptation behavior and the day-to-day update mechanism of traffic flows. The novelty of this thesis is that the proposed approach combines traveler heterogeneity and rationality in a single framework to predict travelers' day-to-day departure time and route decisions, and develops a novel day-to-day dynamic traffic assignment approach. The empirical results obtained from real transportation network, New Jersey Turnpike, confirm that the proposed day-to-day learning and dynamic traffic assignment framework model can successfully capture the significant learning dynamics, demonstrating the possibility of developing a psychological framework (i.e., learning models) as a viable approach to represent travel behavior. The other contributions of this thesis include a novel route choice set generation approach based on stochastic integer programming approach. The proposed methodology takes into account travel time variability and reliability in the transportation network. The path relevance criteria are directly incorporated into the optimization model by minimizing mean travel time, travel time variability and path overlap. Unlike previous approaches in the literature, proposed methodology eliminates the filtering step from the choice set generation and generates paths sets at desired dissimilarity level while minimizing the travel time and variability of these paths. Several case studies show the applicability of the proposed methodology on real transportation networks.