In this paper, a hierarchical neural network architecture for forecasting time series is presented. The architecture is composed of two hierarchical levels using a maximum likelihood competitive learning algorithm. The first level of the system has three experts each using backpropagation and a gating network to partition the input space in order to map the input vectors to the output vectors. The second level of the hierarchical network has an expert using fuzzy ART for producing the correct trend coming from the first level. The experiments show that the resulting network is capable of forecasting the changes in the input and identifying the trends correctly

Meeting Name

IEEE International Conference on Neural Networks, 1994


Engineering Management and Systems Engineering

Keywords and Phrases

ART Neural Nets; Backpropagation; Forecasting Theory; Fuzzy Neural Nets; Gating Network; Hierarchical Neural Network Architecture; Maximum Likelihood Competitive Learning Algorithm; Maximum Likelihood Estimation; Multilayer Perceptrons; Time Series; Time Series Forecasting; Unsupervised Learning

Document Type

Article - Conference proceedings

Document Version

Final Version

File Type





© 1994 Institute of Electrical and Electronics Engineers (IEEE), All rights reserved.

Full Text Link