DefinePK

Abstract

Flight delays pose serious operational and economic challenges to airlines and passengers worldwide. In low-resource settings such as Gilgit-Baltistan and Skardu, delays on domestic routes to remote areas are even more critical due to unpredictable weather, challenging terrain, and limited infrastructure. These regions often lack large-scale data systems and advanced computational resources, making accurate prediction especially difficult. This study explores a practical, low-resource approach to predicting flight delays using only basic operational data. Both classification and regression models were evaluated, and results showed that classification methods, particularly tree-based ensemble models, performed better at identifying potential delays. To support real-world applications, a hybrid implementation pipeline is proposed: a classifier first predicts whether a flight is likely to be delayed, followed by a regressor that estimates the expected delay duration. A conceptual deployment plan is also outlined, suggesting how this system could be used to send early passenger notifications, assist ground staff in smaller airports, and dynamically adjust aircraft turnaround buffers. By focusing on regional constraints and practical deployment, this work offers a feasible and scalable solution for improving flight reliability, operational efficiency, and passenger communication in underdeveloped aviation networks. Future integration into national airline systems could further support policy efforts to enhance service quality in remote regions.