Previsão de séries temporais com máquinas de suporte vetorial

Abstract

This dissertation uses the Support Vector Machine (SVM) technique combining Principal Components and Independent Components analysis in the evaluation of financial time series. This subject is of great interest to researchers, investors and financial institutions that seek to understand the behavior/influence on decision-making in the price market. It is known that the combination of Principal and Independent Components analysis, together with vector support machines can guarantee better results for the context. As a result, it appears that the PCA - SVR, ICA - SV models showed better accuracy when compared to common models, such as the SVR simply. The results of the MAE, MSE, RMSE, R 2 metrics corroborate the applied models in question.