Evolving sparsely connected neural networks for multi-step ahead forecasting

Abstract

Time Series Forecasting (TSF) is an important tool to support decision making. Artificial Neural Networks (ANN) are innate candidates for TSF due to advantages such as nonlinear learning and noise tolerance. However, the search for the best ANN is a complex task that highly affects the forecasting performance. In this paper, we propose a novel Sparsely connected Evolutionary ANN (SEANN), which evolves more flexible ANN structures to perform multi-step ahead forecasts. This approach is compared with a similar strategy but that only evolves fully connected ANNs (FEANN) and a conventional TSF method (i.e. ARIMA methodology). A set of six time series, from different real-world domains, was used in the comparison. Overall, the obtained results reveal the proposed SEANN approach as the best forecasting method, optimizing more simpler structures and requiring less computational effort when compared with the fully connected evolutionary ANN strategy.