Predicting Flight Delays Using Machine Learning

Abstract

Flight delays remain a persistent challenge for the aviation industry, generating high costs and disrupting operations across interconnected networks. Existing monitoring and scheduling tools provide valuable oversight but often lack predictive accuracy and cross-stakeholder coordination, limiting their effectiveness in disruption management. This study develops and evaluates an AI-enabled machine learning framework that integrates operational and meteorological data to forecast delays more reliably. Using U.S. domestic flight records and NOAA weather data for 2024, including a case study at Louisville Muhammad Ali International Airport, we apply classification and regression models to predict on-time performance and delay minutes. Ensemble methods, particularly Random Forest with SMOTE balancing, achieve superior results, detecting delayed flights with 94.7% accuracy and reducing mean absolute error in regression tasks to 4.79 minutes. Beyond technical gains, the framework demonstrates how AI-driven prediction can enhance collaborative decision-making by enabling shared situational awareness across airlines, airports, and air traffic control, strengthening resilience and efficiency in aviation operations.