Temporal prediction using self-organizing multilayer perceptron

In this paper, we apply the self-organizing multilayer perceptron (SOMLP) architecture proposed by Gas for temporal prediction. Our main idea is to divide a data series into several smaller sub-series which are treated as individual functions or signals. Then we can find the tendencies in detail and perform predictions based on the properties of these signals. By using the SOMLP, signals can be clustered and similar sub-series for the underlying prediction are located. The idea is tested by forecasting the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and results are presented.