Schedule Timing Over Carrier Identity: Predicting Flight Arrival Delays at Houston Hub Using Logistic Regression and Random Forest

Flight delays cost airlines and passengers billions every year, yet most prediction models rely on information that only becomes available after a flight has already departed. This paper asks a simpler question: using only what is known before a flight leaves the gate, the time of day, day of week, airline, route distance, and departure airport, can we predict whether it will arrive late? Using the hflights dataset of 223,874 Houston domestic departures from 2011, I built and compared two classification models - Logistic Regression and Random Forest - to predict arrival delays of 15 minutes or more. The results were clear: timing matters far more than the airline. Morning and evening departure periods alone account for roughly 62% of the Random Forest model's predictive power, while no individual carrier contributes more than 3%. Evening flights are nearly four times more likely to arrive late than early morning departures. This is an independent research paper developed following my masters in financial engineering at WorldQuant University. The full Jupyter notebook and dataset are included.