Time series forecasting with generalised additive models

• Autor(a) principal: Abreu Junior, Clony Nunes de
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/34067
• topic: Time series; Regression; Forecast; Prophet; Retail sales; Predict

Time series forecasting is an important area for many types of businesses. Retailers, over time, have frequently sought time series support in their business. This process is currently facilitated by the appearance of applications for users who are not statisticians but who know the business area. Facebook has developed an algorithm called Prophet, which proposes to make estimates based on time series in a quick and accessible way. The work analyses a dataset of 113 retail stores with data of products sold and number of customers, where it is desired to perform forecasts using Prophet and identify how effective the hyperparameter adjustments will be for improving the performance of the estimates projected by the model generated. A preprocessing of data is performed driving adjustments that best suit the use of Prophet. Next, models are created for each of the stores using default parameters and then the results are compared with the models adjusted with hyperparameters. The results presented show the particularities of each shop represented in the specific adjustments of hyperparameters that improve performance, as well as point out the sensitivity of the data to candidate outliers.