The increasing range of ancillary products offered by airlines is making existing, static frameworks obsolete. The changing expectations of customers have created a need for more dynamic and reactive offers. In order to tailor an offer to an individual journey, it is possible to leverage similar journeys and the observed outcomes in a semi-supervised approach.
In this paper, a multi-stage deep learning framework, namely Deep Ancillaries Prediction (DAP), is developed to understand personalised demand for airline ancillaries and improve pricing strategies. DAP aims to solve the overlapping distribution problem and class imbalances observed in real-world airline
datasets. The framework incorporates a contrastive learning module to learn richer feature embeddings and an autoencoder for semi-supervised learning into one framework, and outperforms current ancillary prediction systems. The modules can be trained separately and hence, are suitable for an online learning setting. This framework is designed to be transferable to different
prediction tasks in the airline industry. Significant performance enhancements are attained compared to the current state-of-the-art algorithms.