Authors: Cedric Conol, Rosely Peña, Ren Christian Santos, Mark Lorenze Torregoza

Abstract

The study aims to develop predictive models of domestic flight delays in Ninoy Aquino International Airport (NAIA) through Machine Learning. Flight information such as actual and scheduled departure time, holidays and historical weather were used. The analysis focused only on local flights of one local airline company departing from NAIA from August to December 2018. Feature engineering was performed to generate additional features such as previous aircraft delay and presence of other flights 30 minutes before a scheduled time of departure. Highest accuracy of 95.96% was obtained from the Gradient Boosting classifier model with weather as the most important feature. Another model without using the feature weather condition was tested to determine the internal factors that influence flight delay. 87.5% accuracy was obtained from GBM classifier model with concurrent flight as the top predictor. The study also classified delays into less than 30 minutes, less than 60 minutes and more than 60 minutes. With this, the models were able to determine if a flight scheduled one day away will be delayed with 80% accuracy, flights one week away with 70% accuracy, and two weeks away with 60% accuracy. Additional improvement through feature engineering were also made to raise the performance. Findings of the study can help stakeholders plan for efficient operations to avoid additional variable costs caused by flight delays. Result of the study shows that congestion of flights in NAIA is caused by insufficient runways and accessible airports near Metro Manila.

Note: If you wish to have a copy of the paper (PDF), data and the code used in this project, you may send me a message via Linked in.