Travel time reliability has been regarded as an important factor in travelers route choice decisions.This study explores travelers risk preferences to travel time reliability when they plan a trip.The degree of risk-averse preference is formulated by comparing the on-time arrival probabilities of the least expected travel time path and the selected path under the theory of stochastic dominance.To provide a suitable default value of α (degree of risk-averse preference) for α-reliable shortest path problem in stochastic network, support vector machine is applied to learn and predict the travelers risk preferences by considering variously individual properties (gender, age) and pre-trip information (OD distance, departure time).Large-scale trip records form probe vehicles are utilized to empirical analysis.The tested performances show that the predicted degrees of risk-averse preference by using support vector machine are much closer to the observed data than linear regression.