Predicting Irregular Flight Operations Using a Binary Machine Learning Approach Based on National Meteorological Data

Flight delays are caused by a multitude of external influences as well as revenue driven carrier decisions. Some factors are obvious while others remain inaccessible to the traveling public. Yet knowing of potential flight delays or cancellations in advance can significantly improve passengers’ travel experience and empower them to make informed decisions when flight irregularities occur. We combine a Naïve Bayes - based feature selection method with publicly available meteorological data and flight performance statistics to create a forecasting tool that provides passengers with an improved prediction of potential delays. After promising initial results we optimize our feature selection and weighting, yielding a 66% true positive rate paired with a 66.5% accuracy. This means that 66.5% of our forecasts are correct while the model manages to properly detect 66% of irregular flights. Compared to a probabilistic forecast based on historical data, this represents an improvement of 332% and 436% respectively.

Name thesis:degree_name

Master of Science (MS)

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Aviation

Year

2014

Author dc:creator

• Hellwig, Martin

Repository record dc:identifier

https://commons.und.edu/theses/388

OAI identifier oai:identifier

oai:commons.und.edu:theses-1387