This folder is made available as part of the following research paper in Intelligent Systems with Applications.

Title: Deep Liquid Neural Network for Prediction of Weather-Impacted Flight Delay
Authors: Desmond B. Bisandu, Majuran Chandrakumar and Irene Moulitsas

Citation: Desmond B. Bisandu; Majuran Chandrakumar; Irene Moulitsas. Deep Liquid Neural Network for Prediction of Weather-Impacted Flight Delay. Intelligent Systems with Applications (2026). https://doi.org/10.1016/j.iswa.2026.200656.

Abstract:
In this paper, we developed and examined the prediction of flight delays, with a focus on those caused by weather conditions. Flight delays pose a significant problem for airlines, as the growth of air traffic often leads to financial losses and inconvenience for passengers. Unlike several previous research papers, this study provides a more detailed analysis of delays, distinguishing between weather-related and non-weather-related delays. Deep Learning (DL) models, including the proposed Liquid Neural Networks (LNN), Long Short-Term Memory (LSTM), and Multi-Layer Perceptron (MLP), were employed to make predictions. We compare the LNN model, which has not been previously used in flight delay prediction, with the other two DL models. For this research, we utilised two reliable datasets: flight data from the Bureau of Transportation Statistics and weather data from the Weather Underground site. Our methodological approach is thorough in addressing challenges encountered with the data. One major challenge was managing class imbalance, which can impact prediction accuracy, by using the SMOTE algorithm. The selection of features for training and evaluation was based on mutual information scores and Pearson correlation coefficients. Hyperparameter optimisation was conducted through a systematic approach using grid search and trial-and-error. Models were assessed with standard performance metrics, including precision, recall, F1-score, accuracy, and a confusion matrix. The results show that the LNN model outperformed other DL models in predicting flight delays, especially for the two classes related to flight delays. Overall, all three models yielded excellent results for predicting on-time flights and acceptable outcomes when predicting delays due to non-weather factors. However, the LNN model indicated computational limitations in forecasting weather delays, with notably lower scores for this class and significantly longer training times. Despite these limitations, the LNN model demonstrated engaging performance compared to models from previous research, even outperforming them on some metrics. These findings suggest that the LNN model holds promise for the future of flight delay prediction.


We conducted all our experiments on the Cranfield University High-Performance Computing (HPC) machine called the Crescent 2. This advanced system enabled us to utilise an Nvidia T4 16GB GPU with 8 CPUs. The choice of this GPU was driven by its performance and energy efficiency, making model training faster and more efficient. These two advantages were decisive for carrying out the experiments and obtaining results as soon as possible

## Contents of this repository
This zip folder contains the following files/folders:

- A folder called ‘Weather_Data_Collected’ containing the flight and weather data files collected in CSV format

- A folder called ‘Images’ containing all the figures presented in the paper all jpg format

- A folder called ‘Data_Merged’ containing merged weather and flight data files in CSV format for the various weather data collection timings

- A folder called ‘ResultsFiles’ containing some output files of the results obtained following training and evaluation of the LNN, LSTM and MLP models

- A folder called ‘Testing_Data’ containing different test data files, depending on the timing of the weather data collection used to evaluate the models

- A folder called ‘Training_Data’ containing different training data files, depending on the timing of the weather data collection used to train the models

- A folder called ‘Validation_Data’ containing different validation data files, depending on the timing of the weather data collection used to train the models

- A Jupyter Notebook file called ‘CodingFile_Data_Collection_Data_Preprocessing’ containing the source code for data collection and preprocessing. This file can be run on the CPU of a personal computer, making sure that all the Python libraries are installed.

- A Jupyter Notebook file called ‘CodingFile_Results’ containing the source code for displaying the results obtained during training and evaluation of the models in the form of graphs.

- 3 Python files containing the source code for training and evaluating the LNN, LSTM and MLP models. To run these files, you need to use the high-performance computing platform (Crescent 2) at Cranfield. You will need to import these three python files, the three Sub files and the folders containing the training, testing and validation data files into a high-performance computing platform (Crescent2).

- 3 Sub files containing the code to submit the job to the Crescent2 system for LNN, LSTM and MLP models