In the transportation field, analyzing passengers' travel destinations brings a lot of commercial value. However, research on the passengers' travel destinations is difficult because of its uncertainty. In order to solve this problem, in existing studies, entropy is used to measure the uncertainty of human mobility to describe individuals' travel features, and the spatiotemporal correlation of individual trajectories is taken into account simultaneously, which can not achieve the desired accuracy. Therefore, an algorithm for airline prediction based on improved Markov chain was proposed to predict passengers' travel destinations. First, the distance distribution, site distribution and temporal regularity on history records of passengers' travels were analyzed. Then, the dependence of human mobility on historical behavior and current location was analyzed. Finally, the characteristics of passengers' permanent residence and the exploration probability of new airlines were added into the calculation transition matrix, and an algorithm based on improved Markov chain was proposed and realized to predict passengers' next travels. The experimental results show that the average prediction accuracy of the proposed model can reach 66.4%. Applying in the field of customer travel analysis, airline company can benefit from the research to predict passenger travel better and provide personalized travel services.