Comparing and Combining MLP and NEAT for Time Series Forecasting
Serkan Aras, Anh Anh Nguyen, A. P. White, Shan He · DergiPark (Istanbul University) · 2017
Neural networks are one of the widely-usedtime series forecasting methods in time series applications. Among differentneural network architectures and learning algorithms, the most popular choiceis the feedforward Multilayer Perceptron (MLP). However, it suffers from somedrawbacks such as getting trapped in local minima, human intervention duringthe stage of training, and limitations in architecture design. The aims of thisstudy were twofold. The first was to employ NeuroEvolution of AugmentingTopologies (NEAT), which has many successful applications in numerous fields.In this paper, we applied it to time series forecasting for the first time andcompared its performance with that of the MLP. The second aim was to analysethe performance resulting from the pairwise combination of these methods. Ingeneral, the results suggested that the forecasts from the NEAT algorithm weremore accurate than those of the MLP. The results also showed that pairwise combinedforecasts in general were better than single forecasts. The best forecasts ofall were obtained by pairwise combination of MLP and NEAT.