Dear Vasanthan Devaraj,

You need to build a decision tree to select the features for classification using neural net? What is exactly the application of the decision tree in your scenario?

What do you mean by mapping a decision tree and Neural network?

Mahdieh Zabihimayvan Yes you are right. I need to build a neural network for a tree represented model. I do not need to induce the decision tree using any of its algorithms like ID3 /CART. I will just carry out pattern identification using a tree structure and build a neural network for that model. I will explain the scenario with a simple example. Please find the attached media, it is a tree representation for IRIS data-set. I need to build a neural network for that model. The interconnection between the nodes and the weight assignments are done according to the tree model. The purpose of my work is to make use of the tree structure to understand the concept behind weight assignments and to break down the black-box concept of neural network.

Belcaid Anass I have made corrections to my question for better understanding of the scenario. Hope the edited one will answer your query.

I see now, If I understand correclty you want a neural network to mimic the exact decision tree presented above. Once it's done you want to visualize the layer of that network to get a good understanding of your decision tree.

I suggetst that you train a neural network with a data that represent your decision tree. (i.e) initial data want change but the target of each input will be the result of your decision tree. In that way the neural network will try to mimic exactly the decision tree. Once done, you can visualize the weight of the hidden layer.

Dear Vasanthan Devaraj and Belcaid Anass ,

I agree with Belcaid's suggestion. But, I cannot understand the main goal of having the neural network in this way! If we use decision tree, inputs, and outputs to find the weights of hidden layers, what does such a neural network do in practice? What is its advantage?

Thanks,

Mahdieh

I agree with Mahdieh Zabihimayvan , As I also don't see the point in visualizing the neural network equivalent of a decision tree. But, Vasanthan Devaraj has his owns reasons and goals in seeking this network. Perhaps yould could share with us the main reason.

Mahdieh Zabihimayvan and Belcaid Anass Thank you for your answers. The goal of this task is to explain why a learned network makes a particular decision on a particular test case.