Predictive Modelling of Flight Delay Times Using Machine Learning Algorithms: A Data-driven Approach to Operational Efficiency in Air Travel

Flight delays are a significant concern in the aviation industry, affecting operational
efficiency, customer satisfaction, and overall airline performance. This dissertation aims to
develop a predictive framework using machine learning techniques to forecast departure delay
times based on a variety of features, including airport ratings, carrier performance metrics,
and real-time weather conditions.
The study employs several regression algorithms such as Linear Regression, Decision Tree,
Random Forest, XGBoost, Ridge, Lasso, Support Vector Regression (SVR), and K-Nearest
Neighbors (KNN) to build and evaluate predictive models. Each model is assessed using key
performance indicators-R² score, Mean Absolute Error (MAE), and Root Mean Squared
Error (RMSE)-to identify the most effective approach. Among all, XGBoost emerged as the
most robust and accurate model, offering superior performance in capturing complex, nonlinear
interactions among features.
To ensure real-world applicability, a Gradio-based web application was developed, enabling
users to interact with the model and obtain real-time predictions by inputting relevant flight
and weather parameters. Exploratory Data Analysis (EDA), feature importance ranking, and
correlation studies were also conducted to enhance model interpretability and domain
relevance.