Improved neural network tool: Application to societal forecasting problems

Abstract

Time series forecasting is an important issue for many individuals and enterprises for decision making and planning. Researchers are investigated many time series forecasting problems related to demand, sales, technology, stock, financial forecasting etc. Recent research trend suggests that artificial neural networks are one of the important alternatives to traditional statistical methods. This paper presents improved ANN tools for solving social forecasting problems. This paper presents two multivariate time series forecasting problem for financial organization and agricultural sector. Different models of ANN are developed with varying hidden nodes. Learning rate of the proposed model is adaptive based on the results of past iterations. The experimental result shows that for validation and training, ANN with extended adaptive learning rate outperforms other two variations. Results show that the performance of all ANN increases with a number of hidden nodes up to a certain limit.