Sunday, March 16, 2008

Design Check List for Time Series Forcasting with Neural Network

Designing Neural Network Architecture is an Art not an expert routine. Therefore when designing is based on the trial and error or experience of the developer.
One useful design Neural networks are used in this Time Series Forecasting.
This review gives some basic steps: (Ref: "Computational Intelligence in Time Series Forecasting" chap 3- Neural Network Approach )

Data Preparation:-
Acquisition, Preprocessing , Normalization and structuring of Data.
Determination of training and test datasets.

Network Architecture Determination:-
  • Type of network
  • Number of input, output nodes
  • Number of layers
  • Number of neurons within the layers
  • Interconnection between the neurons
  • Selection of neuron activation function
  • design of the training strategy
  • training algorithm , performance index , training monitoring approch.

1 comment:

Tavita said...

Interesting to know.