Research On Predicting The Growth Of New Passengers In Civil Aviation

With the development and informatization of civil aviation, airlines have accumulated a lot of data, including passenger individual information, querying and booking records, flight information and so on, in their basic operation systems. It is becoming one of the major issues in the field of civil aviation that how to mine more valuable knowledge from the historical operational data. Predicting the growth of passengers, especially new ones, is one of the most meaningful research issues. If we can accurately predict the future value of new passengers based on their rare historical data, it will benefit the airlines to provide personalized travel services to attract potential high-value passengers. Meanwhile, airlines also can classify passengers into different categories and make more accurate marketing strategies to reduce marketing costs according to the predicting results.In this paper, predicting the growth of new passengers is treated as a binary classification problem. Due to the scarcity of historical data of new passengers, predicting their growth by traditional classification methods is not as easy as old passengers who own abundant historical records. To overcome this problem, in this paper I proposed a prediction method to infer the future value of new passengers based on their social relations with old passengers. First, I construct co-travel networks by extracting social relations between passengers from their historical travel records and employ a more reasonable weight measurement model. Then, I respectively generate a series of individual-based and relation-based features to predict the growth of each passenger by employing traditional basic classifiers. Finally, an effective ensemble model is presented to predict the growth of new passengers, which combines the individual-based and relation-based predictions of basic classifiers and improves the final prediction performance.In this paper, I constructed the co-travel networks on a real dataset of passenger travel records of a certain airline. Then the sample data were collected and analyzed according to the analysis result of co-travel networks. By comparing the performance of the traditional classification algorithms, the most appropriate two algorithms were employed as individual-based classifier and relation-based classifier respectively. Finally, the results of several comparison experiments on the dataset demonstrated the signification of this paper and proved that the proposed approach could efficiently predict the growth of new passengers and break the limitation of the scarcity of historical data in the field of civil aviation. Meanwhile, the performance improvements on the two different passenger value measurement models showed that the ensemble model has a very strong generalized ability.