面向不完备数据的民航旅客流失预测模型

Civil aviation passenger loss prediction model for incomplete data

为解决新一代民航旅客服务信息系统中数据缺失不能准确进行旅客流失预测的问题,考虑旅客流失预测任务与旅客缺失数据填补任务具有相关性,提出基于部分距离的SMOTE算法和基于降噪自编码器的多任务深度学习网络模型。基于部分距离的SMOTE算法在数据不完备场景下,对少数类别已流失旅客和准流失旅客进行过采样;设计的多任务深度学习网络模型利用逐层无监督训练和有监督微调,同时进行对缺失数据填补与民航旅客流失预测。实验结果表明,相比传统处理方法,该方法能有效提高数据质量和旅客流失预测的准确性。

To solve the problem that the data loss in the new generation civil aviation passenger service information system causes inaccurate prediction of passenger loss,considering the correlation between the passenger loss prediction task and the passenger missing data filling task,a partial distance based SMOTE algorithm and noise reduction based self-encoding were proposed for multitasking deep learning network model.Through the partial distance-based SMOTE algorithm,in the data incomplete scena-rio,a few categories of lost passengers and quasi-lost passengers were oversampled,and the multi-task deep learning network model was designed to utilize layer-by-layer unsupervised training and supervised fine-tuning.The prediction of missing data and civil aviation passenger loss was carried out.Experimental results show that compared with the traditional processing method,the proposed method can effectively improve the data quality and the accuracy of passenger loss prediction.