:: AForge.NET Framework :: Articles :: Forums ::

Scaling data for Neural Network

The forum is to discuss topics from different artificial intelligence areas, like neural networks, genetic algorithms, machine learning, etc.

Scaling data for Neural Network

Postby chathu03j » Fri Oct 15, 2010 4:15 am

Hi All,

This question is not specific to the AForge framework, however, I would appricaite if someone could give an explanation. I'm trying to use a neural network to forecast the volumes of transaction in financial institutes. We have a forecast made by an existing statistical model, which shows a significant error between the actual volume and the forecast. hence our idea is to use the intial forecast, and other factors such as day of week, whether there's an adjecent holiday, etc... to predict the error, using the neural network.

We need to do this for multiple institutions which have multiple branches for each institution. When analyzing their volumes and the forecast made from the statistical method, we found out that the distribution of error is quite different across institutes. For example, one institue may have around 1000 forecasts in the error (difference between statistical forecast and actual volume) range (0-10) where as another institute may have only 100 in the same range.

In such a case, how can I process the inputs and targets(which are the error values) for thsese network. Will I have to use one network per institution. I was also wondering how this situation is handled in other scenraios such as stock market prediction, where each institute listed in a stock market would have different price scales and volume scales.

I appriciate if someone could give me an idea on this.

Thanks and Regards,
Posts: 10
Joined: Sun Jun 28, 2009 3:43 am

Return to Artificial Intelligence