Enhancing surplus reservation management using artificial intelligence

Frequent flyer programs play a central role in airline revenue management by supporting customer loyalty and using surplus seat capacity. Airlines periodically allocate to these programs surplus seats that remain unreserved through standard commercial channels. A key operational challenge is anticipating whether these seats will be reserved in the short-term, as limited visibility on reservation probability constrains allocation and marketing decisions. This research estimates the probability of surplus seat reservation across multiple flights and markets using classification models based on historical reservation data and flight characteristics. Several machine learning methods are evaluated, with LightGBM identified as the most effective. Three modelling strategies are implemented to enhance predictive performance, and the effects of key explanatory variables are analysed using partial dependence. The best strategy achieves accuracy, recall, and precision above 80%. The resulting probabilities provide short-term decision signals to improve surplus allocation efficiency and profitability.