Flight delay prediction based on LightGBM

With the rapid development of civil aviation industry, the number of flight delays continues to increase. Therefore, various airlines have an urgent need to predict the time of flight delays. The data used is from the data released by the US Bureau of Transportation Statistics (BST). The flight data of Atlanta Hartsfield-Jackson International Airport (ATL) is used as the research data. Data preprocessing is used to clean the data and extract the key data. This paper proposes to use the LightGBM algorithm to predict flight delays, and optimize the parameters of the model through grid search and cross-validation methods. The experimental results show that the LightGBM algorithm outperforms the comparative algorithms in terms of R-Squared, MAE, and training time metrics, exhibiting higher prediction accuracy and shorter training time compared to the XGBoost and GBDT algorithms.