Publicação

Aviation Forecasting Using Machine Learning: Predicting ATC Costs and Fuel Consumption

Ver documento

Detalhes bibliográficos
Resumo:Air Traffic Control costs and Fuel expenses are two of the main expenses of airline companies. Predicting these two variables is essential for the Flight Operations department, as there is a need to calculate the budget for each one and communicate it to the financial department. This master thesis explores the use of machine learning models to predict these expenses, aiming to reduce the dependency on other departments and obtain faster, accurate predictions. The CRISP-DM methodology was adopted to reach the final models for each target variable. Although various machine learning models were trained and tested, Hist Gradient Boosting was the best performing model for each target, however, the results did not meet the expectations. This approach, with the given dataset, performs worse than the company’s existing method for predicting these variables. The main conclusion is that there is a need to engineer existing features and add more relevant ones into the training of the models to fully realize the potential of this approach.
Autores principais:Aguiar, Diogo de Medeiros Cifuentes Nobre
Assunto:Machine Learning ATC Costs Fuel Consumption Predictive Modeling Aviation Forecast SDG 8 - Decent work and economic growth SDG 9 - Industry, innovation and infrastructure
Ano:2024
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso embargado
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL
Descrição
Resumo:Air Traffic Control costs and Fuel expenses are two of the main expenses of airline companies. Predicting these two variables is essential for the Flight Operations department, as there is a need to calculate the budget for each one and communicate it to the financial department. This master thesis explores the use of machine learning models to predict these expenses, aiming to reduce the dependency on other departments and obtain faster, accurate predictions. The CRISP-DM methodology was adopted to reach the final models for each target variable. Although various machine learning models were trained and tested, Hist Gradient Boosting was the best performing model for each target, however, the results did not meet the expectations. This approach, with the given dataset, performs worse than the company’s existing method for predicting these variables. The main conclusion is that there is a need to engineer existing features and add more relevant ones into the training of the models to fully realize the potential of this approach.