Machine learning based analysis and prediction of flight delays in aviation industry

The demand for air transportation has risen as the world's mobility has increased. In addition to the demand for passenger transport, the share of airline transportation, particularly for light and valuable goods, is growing day by day. Increasing demand for both passenger and cargo over the years lead to flight delays, and which has a negative impact on many stakeholders. The fact that the delays cannot be predicted until the last moment conduce to many problems such as congestion at airports, long waiting periods for some passengers or missing connected flights. In addition to the increase in demand for airline transportation, technological developments are enabled to store data easily and using this big data it has become possible to make useful analyzes. Machine learning methods have gained significant momentum in recent years in terms of making these analyzes. In this study, it is aimed to predict aircraft delays by machine learning methods with using detailed flight and weather data together in order to reduce the problems caused by the delays. Accordingly, an international airline company which is operating in the aviation sector is contacted and flight data to be used in the study is obtained through them. Within the scope of flight information, three-year (2016-2017-2018) data are collected such as the planned departure and arrival times of the planes, the departure arrival times of the planes, the airports they travel between, and the exact date information of the trips. In addition to this data, weather data were obtained from online sources. For this, the weather data (temperature, pressure, cloudiness, visibility, extraordinary weather conditions, etc.) at the planned departure /arrival times of the departure and arrival airports are taken into consideration. A data set was created to estimate the status of the flights by matching the airline company's three-year flight data with the weather at the time of aircraft departure and arrival. The created data set was examined and analyzed with Artificial Neural Networks, Random Forests, Extreme Boosting, Light Gradient Boosting and Categorical Boosting, which are among machine learning methods. The performance of these models is evaluated according to various performance metrics based on error matrix and the findings obtained are compared with each other and interpreted.