POTENTIAL HIGH-VALUE PASSENGER DISCOVERY BASED ON SSOMAJ-SMOTE-SSOMIN SAMPLING AND ENSEMBLE LEARNING

Considering highly-imbalanced data and weak correlation between passenger characteristics and value categories of potential high-value passenger, a potential high-value passenger discovery model based on SSOMaj-SMOTE-SSOMin sampling and ensemble learning is proposed. The RFM method was used to label the passenger category. The SSOMaj-SMOTE-SSOMin method was used to resample the imbalanced passenger dataset. The fusion feature selection algorithm (FFS) was used to select the passenger features. Gradient boosting decision tree (GBDT) was taken as the classifier to build a passenger value prediction model to identify potential high-value passengers. Compared with the baseline algorithm, the experimental results on the PNR dataset show that the proposed model achieves better AUC value and F1 value, and can better identify potential high-value passengers.