Modeling flight departure delay distributions

Abstract

In this paper, we construct a probability distribution for flight departure delay durations. The probability distribution is developed by fitting historical departure delay data of an airline to a number of probability distributions. Then, we choose the most fitted model by applying two-stage Genetic Algorithm. In the first stage, the algorithm works for maximizing log-likelihood function, and then we proceed to the second stage by facing the optimization model for minimizing the sum of squared error. The second stage is intended to avoid getting trapped in a local optimal solution which often occurs in log-likelihood approach. Since the historical data show that at a certain time in a day a flight has a bigger probability to be delayed, we also analyze the probability distribution of flight departure delay-time by using a similar approach for distribution of flight departure delay duration. The distributions can be used by the airline for measuring the sensitivity of flight schedules, so that a robust flight schedule can be constructed.