Improvement of Time Series Forecasting Using Genetic-Neural Hybrid Architecture

Time series forecasting has been applied to different applications. Artificial neural networks (ANNs) have widely usage in developing time series forecasting models, too. In this paper, an evolutionary neural network model is proposed to improve the performance of ANNs in time series forecasting. In this way, the genetic algorithm (GA) is used to determine the optimum structure and parameters of ANN, such as the number of hidden nodes, the slope of nodes’ activation function, the values of learning rate and momentum coefficient in hidden and output layers, and the number of input features. To evaluate the effectiveness of the proposed model, foreign exchange rate (FOREX) prediction, as a benchmark application, is performed in this paper. Empirical results show that by using suitable operators for selection and crossover in GA, the mean squared error (MSE) of the proposed evolutionary-connectionist hybrid model reaches 0.0010, which is a better performance compared to some other algorithms.