Potential High Value Passenger Discovery Based On Imbalanced Classification

High-value passenger refers to someone who took a flight with a high frequency and bought ticket with high prices which can bring higher revenue to the airline,are an important source of airline ticket revenue.Potential high-value passenger refers to someone whose current value is lower,but their value will grow rapidly in the future and will soon become high-value passenger.The discovery of potential high-value passenger is an important means for airlines to gain a competitive advantage in the future.Existing researches on potential high-value passenger mainly have the following problems:1.There is redundancy or weak correlation among the available multi-dimensional passenger characteristics;2.the characteristics of data set is presented of highly imbalance classification,compared with the whole passenger group,the potential high-value passenger is only a small sample group;3.The accuracy of the potential high-value passenger discovery model is limited.In response to the above problems,the main works of this article includes:1.Aimed at the problem of redundancy and weak correlation among passengers’ multi- dimensional features,a fusional feature selection method for potential high-value passenger classification is proposed.By fusing the feature selection results of RF,SVM and xgboost,the features that have a significant impact on the value category of passengers are selected.The experimental results show that,compares with single feature selection method,the optimal feature subset selected by this method can effectively improve the experimental results.2.Aimed at the severe classification imbalance problem within the dataset,a data set sampling method based on SSOMaj-ADASYN-SSOMin is proposed.In order to obtain a relatively balanced data set,selective sampling is performed for minority and majority classes respectively.The experiment result shows that the sampling method can effectively improve the performance of potential high-value passenger identification.3.Aimed at the problem of limited prediction accuracy through potential high-value passenger discovery model,a potential high-value passenger discovery model based on SAS-Stacking ensemble learning is proposed,which combined the SSOMaj-ADASYN- SSOMin method with two layers of Stacking ensemble learning.The experimental result shows that the algorithm has further improved the recognition rate of potential high-value passenger.