La predicción es uno de los campos más importantes tanto de la Economía como de la Estadística. El desarrollo de técnicas de predicción automática puede permitir dotarse de información para el futuro e intentar que éste sea lo menos incierto posible para poder anticiparse mejor. Los avances tecnológicos han permitido el desarrollo de un nuevo fenómeno como es el Big Data que permite el tratamiento de grandes volúmenes de datos. En consecuencia, nuevas técnicas como las de Minería de Datos han surgido como alternativas a modelos más clásicos en el campo del automatic forecasting. Así pues, en este trabajo se prevé, por una parte, dar una contextualización de la predicción automática y por otra, de carácter más empírico, llevar a cabo una comparación directa entre algoritmos aplicados a diferentes tipos de datos.

Prediction is one of the most important fields of both economics and statistics. The development of automatic forecasting techniques can enable information to be available for the future and trying to make it as uncertain as possible in order to better anticipate. Technological Innovation has allowed the development of a new phenomenon such as Big Data that enable the processing of large volumes of data. However, this capacity of managing and storing such volume of data need to go with suitable methods, which allow to recognize different patterns and therefore to provide useful information. As a result, new techniques such as Data Mining have emerged as alternatives to more classic models in the field of automatic forecasting. Thus, in this bachelor’s thesis it is envisaged, on the one hand, to provide a contextualization of automatic forecasting, and on the other hand, more empirical in nature, to carry out a direct comparison between algorithms applied to different types of data. In the empirical part, it will be introduced a new algorithm: Prophet. It will be compared with other automatic forecasting algorithms, which come from different statistical models. All of them, except from Prophet, belong to the forecast package of Hyndman. For carrying out the study, it will be use three different time series, which have different characteristics among them such as length, type of trend and seasonality, type of growth… The main purpose of this bachelor’s thesis is giving to the reader a general idea of what is automatic forecasting and discover the amount of applications it has and the amount of algorithms there are. It is important to point out that there are not a winner, but each one works better depending on each data.