How can I create a neural network for data classification using the Neural Network Toolbox?

I want to create and train a neural network which can classify the input data. For example, I would like to classify rocks into one of eight different classes based on the relative concentrations of the different minerals of which these rocks consist.

Answer 1

[Greg Heath]

 Documentation:
     help patternnet
     doc patternnet

 Examples:

      greg patternnet

 Hope this helps.

Greg

Answer 2

[MathWorks Support Team]

A number of neural network types may be trained to perform classification tasks.

For example, this is the textbook job of a perceptron network. It uses one or several hardlimit neurons to divide the input space into 2 or more categories. A single hardlimit neuron splits the input space into 2 halves that are separated by a linear divider (i.e.: a line, a plane, a hyperplane, etc.). Two hardlimit neurons split the space into as many as 4 categories, again separated by linear dividers. Three hardlimit neurons can split the space into as many as 8 categories, and so forth.

The perceptron network has limitations, though: the dividing lines will be linear, and your data may not actually fit the constraints of the classification. Such complication will require more and different neurons. Consult standard texts like Hagan, Demuth, and Beale's Neural Network Design for more information.

The Learning Vector Quantization (LVQ) network is also well-suited for classification applications. It is not bound by the linearity limitations of the textbook perceptron, and it may be more easily adapted to the complications mentioned above. See the attached file "myScript.m" for an example of constructing and training an LVQ network.

[farzad]

This was Awesome help , finally I found a good book , they are reeeeallly few people answering and helping on Neural Networks here , WHY ??????????????????????????????????????

[Greg Heath]

Yeah, the answer is pretty lame. My guess is that the help and documentation is the purview of expert programmers who are not necessarily experts in NNs.