ACER: An adaptive context-aware ensemble regression model for airfare price prediction

Since airlines usually keep their price strategies as commercial secrets and information is always asymmetric, it is difficult for ordinary customers to estimate future flight price changes. However, a reasonable prediction can help customers make decisions when to buy air tickets for a lower price. Flight price prediction can be regarded as a typical time series prediction problem. There are usually two main methods to solve this problem. One is using classical time series prediction methods such as ARIMA, etc. Another is extracting certain features and using regression models. For the latter, sometimes the flight price is context-aware, making it difficult to get an optimized single regression model for the whole price series. Meanwhile, effective context features vary on different air routes and change with time, therefore it is difficult to model context information. In this paper, we propose a context-aware ensemble regression model named ACER which combines different context-aware models and adjusts context features adaptively. Inspired by the idea of bagging and boosting, context features are randomly selected to cluster data efficiently and multiple regression models are trained for data with different contexts. In addition, the context feature list is dynamically adjusted by dropping some irrelevant features. In the experiment on the real data set, our model is compared with the baseline regression model, random forest and classical time series models. The results show that ACER performs much better than the other models.