INTERNATIONAL JOURNAL OF ENGINEERING MANAGEMENT AND
SOCIAL SCIENCES
(An International Peer-Reviewed Multi-Disciplinary Journal)
ISSN : 3139-065X
www.ijemss.com
Scholar ID:
Updated Soon
Scholar URL:
Updated Soon
Scholar ID:
Updated Soon
Scholar URL:
Updated Soon
Scholar ID:
Updated Soon
Scholar URL:
Updated Soon
Flight disruptions, including delays and cancellations, impose substantial economic and operational costs on airlines, airports, and passengers worldwide, making accurate prediction a priority for the aviation industry. This study undertakes a secondary analysis of existing peer-reviewed literature on the application of machine learning to flight disruption prediction, with the purpose of consolidating fragmented findings into a coherent picture of current capability and limitation. Rather than collecting fresh flight data, the paper synthesises reported methodologies, model choices, feature sets, and performance metrics across a range of published studies covering ensemble methods, boosting algorithms, and hybrid architectures. The synthesis shows that tree-based ensemble models, particularly Random Forest and Gradient Boosting variants, consistently outperform simple linear and distance-based classifiers, with reported accuracies often exceeding ninety percent when weather, carrier, and historical delay-propagation features are included. Class imbalance between delayed and on-time flights emerges repeatedly as a major methodological obstacle, addressed unevenly across studies through resampling techniques. The paper concludes that while predictive accuracy has improved steadily, inconsistent evaluation protocols and limited real-time deployment studies remain significant gaps that future research should address.