Integration of SMOTE and Ensemble Models for Predicting Airline Passenger Satisfaction

The high public interest in air transportation has become a polemic for airline companies that are competing to maintain their existence by continuously improving their services. The passenger satisfaction survey data collected has several problems such as unbalanced data, missing values, noise, difficulty finding significant patterns and biased data. Imbalanced class causes the classification results to lean more towards the majority class, this can reduce the performance of the prediction model. SMOTE is one of the over-sampling methods to balance the dataset by increasing the number of samples in the minority class based on k-nearest neighbors to approach the same class. Boosting is a machine learning strategy that combines many very fragile and poor prediction rules to produce very accurate prediction rules. In this study, we conducted a model experiment by integrating the SMOTE and AdaBoost ensembles with the classification algorithm to obtain the best performance metrics. The results showed that the performance of integrating the DT + SMOTE and DT + SMOTE + AdaBoost models produced an accuracy of 91.88%, this performance is superior to the traditional DT model. Significant performance improvements also occur in the integration of NB+SMOTE+AdaBoost and NB+AdaBoost, which is an increase of around 5% compared to NB. However, the application of SMOTE to NB decreases accuracy because SMOTE produces synthetic samples that can disrupt the independence assumption of NB. The results of this study demonstrate the superiority of our proposed method, a robust ensemble learning compared to traditional machine learning classifiers. Both techniques are very efficient in improving classification capabilities, especially in cases of complex and imbalanced data. AdaBoost, Customer satisfaction prediction, Data mining, Ensemble learning, Imbalanced data, SMOTE.