I have developed six Neural Networks for time series forecasting, more precisely for prediction of future values of a commodity prices. The data set I am using is flour prices over 100 months for three different cities. I am looking for a code written in MATLAB, of a Genetic Algorithm, or any other method of AI, that would optimize the architecture of the neural networks.
## Deliverables
What should be hand in to me, is the code of the GA, preferably written in MATLAB. Probably a specialized MATLAB toolbox for GA's might be needed. I am suggesting the Gatbx which can be downloaded from [login to view URL]~gaipp/ga-toolbox/ or GEATbx that I work on. The neural networks I have produced are MultiLayerPerceptrons with Backpropagation. There are two kinds of them. The first is for Univariate analysis of the prices, which means the NN forecasts future values for a city, based only on previous values of this city. The NN's I have are 2-2-1, 4-4-1, 6-6-1, 8-8-1. The first number represents the inputs, the second represents the nodes of the hidden layer and the output is always one. I want the GA to be able to recognize which architecture is better based on the criterion MSE (Mean Square Errors) which is actually the performance function of the NN's. So the goal for the GA would be to find the optimum number of input and hidden nodes. The second kind includes two NN's built for multivariate analysis, where for each city, predictions are based on data from three cities. The networks are: 6-6-1 and 8-8-1. In this case only the optimum number of hidden nodes is asked for each NN. Complete copyrights to all work purchased.
## Platform
MATLAB 6.1 wich runs under Win XP
## Deadline information
If time is not enough, I am thinking of getting an extention for my project, so an extra week might be available.