Application of Predictive Modelling to Determine Factors Influencing Flight Delays at Jomo Kenyatta International Airport

Abstract

o improve decision making accuracy in any given airport requires selecting imp ortant variables that can be used in building a predictive model. The choice of appropriate independent variables will improve the model precision . However, the choice of the independent variables depends on the data that is recorded by the airlines and ai rports on the flights. The airport managers would want to understand the key factors behind flight delays. It is thus important to make a comparison on the mostly used factors in many modelling studies of flight delays and the factors that influence flight delays at Jomo Kenyatta International Airport ( JKIA ) . The factors mostly used are the weather and the flight features . The factors available at JKIA include; the day of the flight (that is, Monday to Sunday), the month (that is, January to December), the airline, the flight class (that is, domestic or international), season (that is, summer (March to October) or winter (October to March), capacity of the aircraft, flight ID (tail number) and whether the flight had flown at night or during the day . The data used was obtained from Kenya Airports Authority for the JKIA flights for the period from March 2017 to March 2018. The analysis was done using R - Gui statistical software. Descriptive statistics were generated to give a general overview of how the above fa ctors influenced flight delays at JKIA. Logistic model was then fitted to demonstrate how the factors could be applied in predicting flight delays. This model was also used to extract the significant factors in predicting flight delays. The selected factor s were also compared on the performance they yielded in modelling as compared to features which had been used in other studies. The results revealed that the significant factors were days of the week, months, flight class and capacity. Modelling using thes e factors yielded models with average F1 score of 76.95%. This was better performance when compared to results from another study that used predictive features such as: the previous aircraft arriving late, weather, and departure time and ac hieved an average F1 score of 58.7 %. Another study predicted airline delays using flight departure times, and weather conditions. Their prediction algorithms achieved F1 score of 56 . 6 % . This shows that the factors that influence flight delays at JKIA improves the performan ce of predictive models of flight delays.