Smart Prediction of Weather-Induced Flight Delays Applying Deep Learning

Abstract

Flight delay prediction is a critical challenge in modern aviation, affecting passenger satisfaction, airline efficiency, and airport operations. This paper presents SkyPulse AI, a hybrid machine learning framework that integrates XGBoost and an Artificial Neural Network (ANN) through soft voting to improve classification robustness. A regression model independently estimates delay duration in minutes when a delay is predicted. All three models are deployed through a Flaskbased web application that provides real-time predictions, SQLite-backed prediction history, an interactive analytics dashboard, and automated email notifications via the SendGrid API. Evaluated on a flight-and-weather dataset comprising ten input features, the hybrid model achieves 91.5% accuracy, outperforming standalone XGBoost (89.0%) and ANN (88.2%) baselines. Results confirm that combining complementary model strengths through soft voting yields stable probability estimates and reduces prediction variance.