Loading...
Publication
Modelação e previsão da procura turística na Ilha do Sal – Cabo Verde: modelo SARIMA versus rede neuronal artificial
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.82 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O objetivo principal deste trabalho é a análise da série temporal "Número de dormidas mensais nos estabelecimentos turísticos da ilha do Sal - Cabo Verde” entre 2000 e 2018 e demonstrar que, no quadro dos modelos quantitativos, estimativas e previsões fiáveis para o comportamento da procura turística são preferencialmente obtidas com modelos estatísticos de previsão adequadamente especificados e testados, e com redes neuronais artificiais que permitem prever um passo à frente. Para tal, iniciou-se a investigação com
uma revisão da literatura e análise de dados que possibilitou conhecer: a dinâmica do turismo mundial; o mercado emissor europeu; o desenvolvimento turístico da ilha; as características da série temporal; modelos de previsão e tendências da procura turística.
Seguidamente implementaram-se diferentes estruturas de modelos de previsão. Os resultados obtidos mostram que, a nível individual, a ilha se encontra na fase de Desenvolvimento; a nível de competitividade, a ilha está estagnada dentro da fase de
Exploração; e que o seu Índice de Desenvolvimento Turístico deverá crescer em 48% para
entrar na fase de Envolvimento. Quanto aos modelos de previsão, obtiveram-se: o
modelo SARIMA(2,1,0)(0,1,1)[12], com uma acurácia medida pelo MAPE igual a 6,77%; o modelo de redes neuronais do tipo RNAR(12,1,7)[12], com um erro de 5,61% e o método de Holt-Winters que produziu um modelo com uma precisão de 7,94%. Todos esses modelos têm alta precisão, com destaque para a rede neuronal, apesar dos dados da série não estarem adaptados à Lei de Benford. Porém, o proposto Algoritmo de Atribuição do Erro, traz melhorias ao resultado do modelo SARIMA, com uma precisão de 4,98%. Esta tese pretendeu contribuir para mostrar o potencial dos modelos estatísticos de previsão e da aplicação das redes neuronais artificiais para a previsão do número de dormidas mensais na ilha do Sal. Também se avalia a precisão das previsões de cada modelo e compara os seus diferentes desempenhos.
The main objective of this work is the analysis of the time series "Number of monthly overnight stays in tourist establishments on the Island of Sal- Cape Verde" between 2000 and 2018. In the framework of quantitative models, one can obtain reliable estimates and forecasts for the behavior of tourism demand, using statistical forecasting models adequately specified and tested, and artificial neural networks that allow one step ahead predictions. With this goal, the research began with a literature review and data analysis that allowed do understand: the dynamics of world tourism; the European outbound market; the island's tourism development; the time series characteristics; the forecasting models and tourism demand trends. Then, different forecast model structures were implemented. The results obtained show that, at the individual level, the island is in the Development stage; at the competitiveness level, the island is stagnated within the Exploration stage; and that its Tourism Development Index should grow by 48% to enter the Involvement stage. As for the prediction models, the following were obtained: the SARIMA(2,1,0)(0,1,1)[12] model, with an accuracy measured by MAPE equal to 6.77%; the neural network model of type RNAR(12,1,7)[12], with an error of 5.61% and the Holt Winters method that produced a model with an accuracy of 7.94%. All models show a high accuracy, with the neural network standing out, despite the series data not being adapted to Benford's Law. However, the proposed Error Assignment Algorithm brings improvements to the result of the SARIMA model, with an accuracy of 4.98%. This thesis intended to contribute to show the potential of statistical forecasting models and the application of artificial neural networks to forecast the number of monthly overnight stays in the Island of Sal. The forecasting accuracy of each model was also evaluated, and their different performances were compared.
The main objective of this work is the analysis of the time series "Number of monthly overnight stays in tourist establishments on the Island of Sal- Cape Verde" between 2000 and 2018. In the framework of quantitative models, one can obtain reliable estimates and forecasts for the behavior of tourism demand, using statistical forecasting models adequately specified and tested, and artificial neural networks that allow one step ahead predictions. With this goal, the research began with a literature review and data analysis that allowed do understand: the dynamics of world tourism; the European outbound market; the island's tourism development; the time series characteristics; the forecasting models and tourism demand trends. Then, different forecast model structures were implemented. The results obtained show that, at the individual level, the island is in the Development stage; at the competitiveness level, the island is stagnated within the Exploration stage; and that its Tourism Development Index should grow by 48% to enter the Involvement stage. As for the prediction models, the following were obtained: the SARIMA(2,1,0)(0,1,1)[12] model, with an accuracy measured by MAPE equal to 6.77%; the neural network model of type RNAR(12,1,7)[12], with an error of 5.61% and the Holt Winters method that produced a model with an accuracy of 7.94%. All models show a high accuracy, with the neural network standing out, despite the series data not being adapted to Benford's Law. However, the proposed Error Assignment Algorithm brings improvements to the result of the SARIMA model, with an accuracy of 4.98%. This thesis intended to contribute to show the potential of statistical forecasting models and the application of artificial neural networks to forecast the number of monthly overnight stays in the Island of Sal. The forecasting accuracy of each model was also evaluated, and their different performances were compared.
Description
Keywords
Turismo Série temporal Modelo SARIMA Holt-Winters Rede Neuronal Artificial Cabo Verde Ilha do Sal Tourism Time series SARIMA Artificial Neural Network Sal island
Citation
Neves, Gilberto A. - Modelação e previsão da procura turística na Ilha do Sal – Cabo Verde [Em linha]: modelo SARIMA versus rede neuronal artificial. [S.l.]: [s.n.], 2022. 371 p.