基于机器学习的航空乘客满意度预测

Machine Learning-Based Prediction of Air Passenger Satisfaction

摘要: 应用Transformer以及其他机器学习算法对航空乘客满意度进行综合评价，分析航空乘客对航空服务质量的满意度及其影响因素并对其满意度水平进行预测研究，为提高航空服务质量提供参考。首先通过文献调研发现，大多数研究集中在2015年美国航空乘客满意度数据集上。为避免与现有研究重复，本研究从kaggle平台选取了一个与2015年美国数据集特征相同且样本量较大的数据集，通过合并这两个数据集，进行一项更深入的分析。然后通过现有特征构建两两特征间相加和相乘的新特征，通过用一个较为简单的模型LightGBM来对比看哪种特征处理方式获得的新特征最佳，最后对比多个机器学习模型，确定最优模型。结果显示，Transformer在准确率、召回率的三个指标均优于其他算法，其中Accuracy：0.9592，AUC：0.9604，Precision：0.9573，Recall：0.9604，F1-score：0.9586。在线登记是航空服务质量满意度的重要影响因素，航空公司应进一步加强以上方面的培训建设。

Abstract: Transformer and other machine learning algorithms are used to comprehensively evaluate the satisfaction of air passengers, analyze the satisfaction of air passengers with aviation service quality and its influencing factors, and predict the level of air satisfaction, so as to provide reference for improving the quality of aviation services. First of all, through literature research, it is found that most of the research focuses on the 2015 air passenger satisfaction dataset in the United States. In order to avoid duplication with existing studies, this study selected a dataset with the same characteristics as the 2015 US dataset and large sample size from the Kaggle platform and combined the two datasets for a more in-depth analysis. Then, through the existing features, a new feature is constructed to add and multiply between two features, and a simple model, LightGBM, is used to compare which new feature processing method is the best. Finally, multiple machine learning models are compared to determine the optimal model. The results show that Transformer is better than other algorithms in terms of accuracy and recall, Accuracy: 0.9592, AUC: 0.9604, Precision: 0.9573, Recall: 0.9604, and F1-score: 0.9586. In the decision-making of satisfaction. Online boarding is an important factor in aviation service quality satisfaction, and airlines should further strengthen the training construction in the above aspects.