Modelos preditivos para previsão de crises financeiras

Abstract

Computer Science is one of the areas of knowledge that deals with the development of methods and techniques for building predictive systems.A predictive system consists of recognizing the existing patterns in the data to predict and anticipate future trends. In this context, the aim of this work is to analyze data from the financial area, such as business cycle movements, using machine learning techniques for forecasting purposes. Building predictive models in the financial area is a challenge for governments and institutions, since it involves human factors and international events. On the other hand, financial crises are rare events, and therefore difficult to predict. However, they have a considerable impact, which deserves to be studied in order to be predicted and remedied. This work was based on a systematic review, from which some related projects were studied. The models proposed by the related works for predicting from financial data for portfolio management problems, bankruptcy risk, stock exchange collapse and business cycles. The models reported are studies and researches, which still need improvement. In the absence of a definitive approach to the problem of forecasting financial crises, it is proposed in this work to investigate machine learning and in-depth learning methods applicable to it. In terms of data sets, there are few available for this purpose. The dataset that will be used in this work is the database called the Jordà-Schularick-Taylor Macrohistory. As software tools, Python and its libraries, widely used in Data Science, were selected. [resumo fornecido pelo autor]