Flight Delay Prediction Using Machine Learning

Abstract

Predicting a delay class or value is the most common approach to the problem of predicting flight delays. Probabilistic delay predictions on an individual flight basis, on the other hand, provide insight into the uncertainty of delay predictions and can be of great benefit to the aviation industry. As a result, in order to predict flight delays at a European airport, this study employs Mixture Density Networks and Random Forest regression, two probabilistic forecasting algorithms. With a Mean Absolute Error of less than 15 minutes, the algorithms estimate the distribution of arrival and departure flight delays accurately. In a probabilistic flight-to-gate assignment problem, we incorporate these probabilistic predictions to demonstrate the value of the estimated delay distributions. Increasing the robustness of flight-to-gate assignments is the goal of this problem. When compared to a deterministic flight-to-gate assignment model, our proposed flight-to-gate assignment model reduces the number of conflicted aircraft by up to 74% when probabilistic delay predictions are taken into consideration. Overall, the findings demonstrate the value of probabilistic forecasting for stable airport operations optimization