Optimizing Flight Delay Predictions by Integrating Hybrid Deep Learning Architectures with Particle Swarm Optimization

Karimpour, Maryam; Shayannia, Seyed Ahmad -; Rajabzade Ghatari, Ali -; Iranban, Seyed Javad -

doi:10.22034/jast.2025.532838.1231

Optimizing Flight Delay Predictions by Integrating Hybrid Deep Learning Architectures with Particle Swarm Optimization

Articles in Press

Document Type : Original Article

Authors

Maryam Karimpour ¹

Seyed Ahmad - Shayannia ²

Ali - Rajabzade Ghatari ³

Seyed Javad - Iranban ⁴

¹ Department of Management and Economics, SR.C., Islamic Azad University, Tehran, Iran.

² Department of Industrial Management, Fi.C., Islamic Azad University, Firoozkooh, Iran.

³ Department of Management and Economics, Tarbiat Modares University, Tehran, Iran.

⁴ Department of Management, Shi.C., Islamic Azad University, Shiraz, Iran.

10.22034/jast.2025.532838.1231

Abstract

Flight delays significantly affect airline operations and passenger satisfaction. While deep learning models such as long short-term memory (LSTM) and gated recurrent units (GRU) are promising for prediction, their performance is highly sensitive to hyperparameter configuration.
This study proposes a novel hybrid deep learning framework integrated with particle swarm optimization (PSO) to enhance prediction accuracy. We develop a model that synergistically combines GRU and LSTM layers to capture short-term and long-term temporal dependencies in flight data. Importantly, the PSO meta-heuristic algorithm is employed to automate the optimization of key hyperparameters, including the number of units, dropout rate, learning rate, batch size, and training periods. The model was trained and tested on a comprehensive dataset of US airports, including features such as weather, air traffic volume, and operational details. Our results show superior performance with 98.24% accuracy, 94.00% precision, 96.00% recall, and 95.00% F1 score, significantly outperforming the baseline LSTM/GRU models and other methods reported in recent papers. This research emphasizes the critical role of systematic hyperparameter tuning and provides a robust and accurate tool for flight delay prediction, with potential applications in airline resource planning and real-time delay management.

Keywords

Flight Delay Prediction

Deep Learning

Parameter Optimization

Particle Swarm Optimization (PSO)

Subjects