| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 3.71 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Este estágio teve como principal objectivo encontrar uma alternativa ao método de previsão utilizado pela empresa Vodafone no que diz respeito às suas linhas de apoio aos seus clientes.
Os dados referem-se ao número de chamadas atendidas diariamente de diferentes linhas de apoio e portanto estas linhas podem ser tratadas como séries temporais. Pretende-se encontrar quais os melhores modelos que se ajustam às series temporais diárias observadas para as várias linhas de apoio a clientes da referida operadora móvel e depois usaremos estes modelos para prever os valores futuros para estas séries temporais. Um dos objectivos é avaliar a qualidade de ajustamento bem como o poder preditivo destes modelos.
O método de previsão sugerido é a clássica Metodologia de Box-Jenkins que se baseia nos modelos de séries temporais lineares. Grande parte das nossas séries temporais, apresenta uma forte componente sazonal diária e portanto os modelos que iremos aplicar são os chamados modelos lineares multiplicativos não estacionários sazonais que designam-se por SARIMA(p,d,q)x(P,D,Q)S. Ao fazermos uma análise cuidadosa destas séries temporais, por vezes estas mostranos algumas não linearidades e desta maneira devemos aplicar os modelos GARCH para os resíduos obtidos dos modelos lineares SARIMA, para explicar o elevado grau de volatilidade presente em algumas séries temporais.
No final deste relatório, apresento detalhadamente um caso prático referente a uma linha de atendimento que se encontra no activo na empresa. Tendo em conta o modelo que ajustei para esta linha, pretende-se prever os valores desta para os meses de Novembro e Dezembro de 2011 e para o mês de Janeiro de 2012 e comparar esses valores com os valores reais dessa linha e também pelo método utilizado na Vodafone, para depois retirarmos conclusões acerca dos mesmos. Posteriormente, realizamos o mesmo procedimento mas agora para os meses de Junho, Julho e Agosto de 2012.
Uma vez que até a este momento só faço o retrato de uma linha, também irei aplicar resumidamente estes métodos a outras linhas que faltam considerar, para depois tirarmos uma conclusão final acerca dos métodos apresentados. Os resultados vão-nos mostrar claramente a superioridade do método Box-Jenkins sobretudo para períodos curtos de previsão em relação ao método aplicado pela Vodafone.
The main objective of this training report is to find an alternative method of prediction to the methods currently in use by the Vodafone company for customer helpline telephone traffic. Data correspond to the daily number of telephone calls received on several different helplines and therefore are treated as time series. We hope to find models which fit best to the observed daily time series of several different customer helpline traffic, and then use these models for prediction of future values of these time series. The objective is to assess the quality of fit as well as the predictive power of these models. The method of prediction which we suggest is the classical Box and Jenkins Methodology based on linear time series models. Almost all the observed time series show strong weekly seasonal components and therefore the models which we employ are the seasonal non-stationary linear multiplicative models designated by SARIMA(p,d,q)X(P,D,Q)S. The careful analysis of the time series shows that there is non-linearity and we employ GARCH models to the residuals obtained from the linear SARIMA models to explain the high degree of volatility in some of the time series. At the end of this report a case study will be presented referring to a hotline which is in active in the company. Based on the adjusted model, the predicted values of number of calls for this line in the months of November, December 2011 and January 2012 were compared with the real values (observed) and also with the method used in Vodafone. Afterwards, the same methodology was applied for the months of June, July and August 2012 for this line, as well as in other lines and conclusions were drawn about the presented methods. The results will clearly show the superiority of the Box and Jenkins method particularly for short term predictions over the methods employed by the Vodafone.
The main objective of this training report is to find an alternative method of prediction to the methods currently in use by the Vodafone company for customer helpline telephone traffic. Data correspond to the daily number of telephone calls received on several different helplines and therefore are treated as time series. We hope to find models which fit best to the observed daily time series of several different customer helpline traffic, and then use these models for prediction of future values of these time series. The objective is to assess the quality of fit as well as the predictive power of these models. The method of prediction which we suggest is the classical Box and Jenkins Methodology based on linear time series models. Almost all the observed time series show strong weekly seasonal components and therefore the models which we employ are the seasonal non-stationary linear multiplicative models designated by SARIMA(p,d,q)X(P,D,Q)S. The careful analysis of the time series shows that there is non-linearity and we employ GARCH models to the residuals obtained from the linear SARIMA models to explain the high degree of volatility in some of the time series. At the end of this report a case study will be presented referring to a hotline which is in active in the company. Based on the adjusted model, the predicted values of number of calls for this line in the months of November, December 2011 and January 2012 were compared with the real values (observed) and also with the method used in Vodafone. Afterwards, the same methodology was applied for the months of June, July and August 2012 for this line, as well as in other lines and conclusions were drawn about the presented methods. The results will clearly show the superiority of the Box and Jenkins method particularly for short term predictions over the methods employed by the Vodafone.
Descrição
Tese de mestrado em Estatística, apresentada à Universidade de Lisboa, através da Faculdade de Ciências, 2012
Palavras-chave
Previsão Série temporal Metodologia Box-Jenkins SARIMA GARCH Teses de mestrado - 2012
