Machine Learning Methods Benchmarking for Predicting Flight Delays: An Efficiency Meta-Analysis

Predicting delays in commercial flights is an increasing challenge due to rising air traffic demand, which generates additional costs and operational complexity. This study synthesizes and evaluates machine learning approaches for flight delay predictions, aiming to identify the most accurate prediction logic and assess the role of sample size in model performance. A systematic literature review was conducted, followed by a meta-analysis of 1077 studies published between 2015 and 2025. The studies were classified by prediction logic (binary classification or regression) and evaluated in terms of model effectiveness using Data Envelopment Analysis and Tobit regression to determine the influence of explanatory variables. The results show that binary classification approaches achieved higher average accuracy than regression models did, with confidence intervals validating their relative effectiveness. Furthermore, findings indicate that the use of more complex models does not guarantee improved predictive performance, suggesting that researchers should prioritize robust variable selection rather than constantly adopting increasingly complex methods. This work provides a comprehensive overview of machine learning methods for flight delay predictions and highlights implications for optimizing airport operations and enhancing passenger experience through the adoption of more reliable predictive strategies.