Optimizing the artificial neural network parameters using a biased random key genetic algorithm for time series forecasting

Abstract Artificial Neural Networks (ANN) is one of the most used methods in time series forecasting. Mostly, it is hard to determine the design and weight parameters of ANNs by experience. For this reason, ANN optimization is considered to find the best network design and parameter sets. In this study, a novel hybrid algorithm based on biased random key genetic algorithms is proposed for ANN optimization. The algorithm, which is named as BRKGA–NN, determines the number of hidden neurons, bias values of hidden neurons and the connection weights between nodes. To test the performance of the BRKGA–NN, the algorithm is compared with genetic algorithm based ANN, ANN with back-propagation, Support Vector Regression and Autoregressive Integrated Moving Average on some of the most known time series datasets. According to the results of forecasts, BRKGA–NN algorithm can produce better forecasts than the compared methods. In addition to the time series datasets, the results and comparisons of the forecasts with the proposed algorithm using real-life data are also presented.