Abstract

This thesis presents the development and validation of a machine learning model using Long Short-Term Memory (LSTM) networks to forecast air traffic passenger volumes at international, high-traffic airports. By leveraging historical passenger data, the study seeks to provide accurate predictions that can enhance operational planning and resource allocation at the airport. The implementation was done in Python, using historical passenger data from San Francisco International Airport (SFO) from 2005-2018. Data preprocessing steps involved grouping and summing passenger counts for each activity period. The data was normalized to a range of 0 to 1, which is crucial for preventing instability and improving the gradient performance of deep neural networks. The dataset was then split into 66% training and 34% testing sets, a common practice for time series forecasting. The LSTM model was built to use a look-back period of 1 to predict future values. The model architecture included an LSTM layer with 4 units and a Dense output layer, and it was trained over 100 epochs using the Mean Squared Error (MSE) loss function and Adam optimizer. Training and validation loss values were tracked, and while the model demonstrated good learning behavior on the training set, the validation loss was higher, suggesting a degree of overfitting. The trained LSTM model was evaluated using MSE and Root Mean Squared Error (RMSE) metrics. For the training set, the model achieved an RMSE of 262,064.79 passengers, while for the test set, the RMSE was 396,732.79. Although the model showed some overfitting to the training data, it generalizes reasonably well to unseen data. The model’s predictions were then inverse transformed to the original passenger count scale, allowing for a direct comparison to actual values. Finally, the predicted values were visualized alongside the actual data. The model captured the overall trend and seasonal patterns of passenger volumes at SFO, with relatively accurate predictions on the training data and reasonable performance on the test data. While the model showed challenges in generalizing perfectly to unseen data, the results provide valuable insights for improving airport operational efficiency through more informed resource allocation and planning.

Publication Date

11-19-2024

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Graduate Programs & Research

Advisor

Sanjay Modak

Advisor/Committee Member

Khalil Al Hussaeni

Campus

RIT Dubai

Download

Share

COinS