DECOMPOSITION AND FORECASTING TIME SERIES IN BUSINESS ECONOMY USING PROPHET FORECASTING MODEL

There are many methods of forecasting, often based on the specific conditions of the given time series which are frequently the result of research in scientific centres and universities. Nevertheless, there are also models that were created by scientists in a particular company, examples may be Google or Facebook. The latter one has developed one of the latest Prophet forecasting models published in 2017 by Taylor & Letham. This model is completely new and so it is appropriate to subject it to further research, which is the topic of this article. To accomplish this research objective, the aim of this work is to identify seasonal trends in revenue development in a selected e-commerce segment based on the assessment of the applicability of the Facebook Prophet forecasting tool. To accomplish this goal, the Python Prophet is decomposed with a subsequent two-year forecast. Accuracy of this model is measured by RMSA and coverage. The e-commerce subject selected is active primarily in the field of sales of professional outdoor supplies and organizing outdoor educational courses, seminars and competitions. It is clear from the prediction that the e-commerce entity shows a strong sales period with the beginning of the spring season and then, due to the summer, decline, until the pre-Christmas period. The subject has little growth potential and a new impetus is needed to increase sales and thus restore the growth trend. It has been confirmed that Prophet is a suitable tool for debugging seasonal tendencies.