Industrial Resilience: Optimising Airline Resilience with Graph Mining - DIAL Newsletter Summer 2019
Our collaborative project with Boeing - APEMEN (Airline Performance and Disruption Management across Extended Networks) deals with the analysis of historical operational data from airlines to find the root causes of disruptions and their impact on the airline schedules. The aim of the project is to ultimately develop recovery solutions to bring the disrupted schedules back to normal state.
Flight delays are caused by a number of causes like weather issues, late passengers, late/sick crew members and aircraft issues etc., which lead to deviations from the planned schedules of the airlines. These flight delays have huge impact on the airlines in terms of money and reputation, on passengers in terms of time and money, nations in terms of economic losses and on the environment in terms of air pollution, disruptions cost up to $60 billion per year or about 8% of the worldwide airline revenue to the airlines and their customers .
Prediction of flight delays is crucial in the decision-making process and can help in the recovery solutions. Nowadays, machine learning techniques like logistic regression, SVM, random forests, Naïve Bayes, linear regression and deep learning are being used to predict the flight delays, airport delays and disruptions. These techniques help to foresee the disruptions in the airline schedules which help to timely prepare for the recovery solutions and help to avoid the propagation of delays to other connected flights. Thus machine learning techniques help in reducing the economic losses by avoiding large disruptions and can also help in improving the marketing decisions.
For example, in APEMEN we developed an analysis framework which models the historical data in terms of a flight connection network where flights are nodes, and passenger, crew or tail connections are links. The network is then transformed to delay networks. The subsequent application of community detection algorithms and frequent subgraph mining helps us find the root-causes of delays and their frequency. The framework gives us important statistical insights to improve airline operations.