Artif Intell Rev DOI 10.1007/s10462-011-9275-1 Forecasting models for prediction in time series Ot ´ avio A. S. Carpinteiro · Jo ˜ ao P. R. R. Leite · Carlos A. M. Pinheiro · Isaías Lima © Springer Science+Business Media B.V. 2011 Abstract This paper presents the study of three forecasting models—a multilayer percep- tron, a support vector machine, and a hierarchical model. The hierarchical model is made up of a self-organizing map and a support vector machine—the latter on top of the former. The models are trained and assessed on a time series of a Brazilian stock market fund. The results from the experiments show that the performance of the hierarchical model is better than that of the support vector machine, and much better than that of the multilayer perceptron. Keywords Kernel-based models · Neural models · Hierarchical models · Artificial intelligence · Financial time-series forecasting 1 Introduction Models to predict values in time series are necessary in many areas of knowledge. For instance, these forecast models are necessary to predict temperature values, prices of shares in the stock market, and electrical energy consumption. The search for models which yield more accurate predictions has advanced the research both in the areas of kernel-based models (Cao 2003) and of neural models (Zhang and Kline 2007). Particularly, neural models should include some kind of mechanism to analyse the time series in order to produce reliable predictions. Time windows (Kangas 1994) and time integrators (Chappell and Taylor 1993) are by far the most employed mechanisms. O. A. S. Carpinteiro (B ) · J. P. R. R. Leite · C. A. M. Pinheiro · I. Lima Research Group on Systems and Computer Engineering, Federal University of Itajub ´ a , Itajub ´ a, MG 37500–903, Brazil e-mail: otavio@unifei.edu.br J. P. R. R. Leite e-mail: joaopaulounifei@gmail.com C. A. M. Pinheiro e-mail: pinheiro@unifei.edu.br I. Lima e-mail: isaias@unifei.edu.br 123