We develop a personalized system to modify individual travel behaviors by offering personalized incentives. Individual preferences are learned to provide personalized incentives so that the promoted alternative is likely accepted. Using knowledge from control theories and state estimation, we model travelers’ choice-making behaviors with the random utility theory and responses from the individuals are mined by a particle filter for learning individual preferences to promote sustainable behaviors. The discrete nature of travel behavior naturally leads to limited observability. We overcome this problem by designing a measurement function from which additional information can be solicited. Additionally, the inherent trade-offs between factors that affect travel choices result in an infinite set of solutions. We thus propose two solutions: (1) the divide and conquer strategy in which a multi-dimensional conditional probability function is proposed; and (2) use of domain knowledge to restrict that preference values fall in certain ranges and are consistent with certain distributions. The performance of preference learning with these two solutions applied is shown via simulation tests and an online experiment involving human participants. For departure time choices, we show an average acceptance ratio of 0.68 for all participants when being promoted with alternatives with personalized incentives. We also show that changes in individual departure time choices will lead to 48% reduction in total travel time on a simple transportation network.