A classic regression dataset for predicting airline delays. We have processed the original sources to match the dataset used in several papers on large-scale GP and kernel regression. The dataset first appeared in the learning community following the 2009 data expo.

The goal is to predict the delay of airplane flights starting from basic information about those flights (time of day, distance, age of the plane, etc.). The data is already mostly cleaned up (no missing data), but there are a few outliers, and the format of some of the columns is not ideal for learning: the ArrTime and DepTime columns are in 'hhmm' format stored as a floating point numbers.

The notebook connected to this dataset shows simple preprocessing which can be performed for cleaning the data - as used in papers such as

• Variational Fourier Features for Gaussian Processes
• Kernel methods through the roof: handling billions of points efficiently