2-Class Problem with patternnet

1 view (last 30 days)
Hej, I have a problem. Using the NN toolbox a neural network shall be trained to recognize a two class problem. I used the default settings ( dividerand , 10 hidden neurons, divide radio 0.7, 0.15, 0.15) and my input is a 3xn matrix and my target is a 2xn matrix ([0; 1] for class one and [1; 0] for class two for each sample), where n=21000. the ratio of the classes are about 3:2.
Why my confusion matrix looks like:
Thank you very much!!

Accepted Answer

Greg Heath
Greg Heath on 13 Dec 2013
Edited: Greg Heath on 14 Dec 2013
Class 1 target should be [1;0] and class 2 should be [0;1]
N1 = 13840, N1trn = 9421, N1val = 2044, N1tst = 2015
N2 = 7895, N2trn = 5542, N2val = 1162, N2tst = 1191
All inputs except one class2 validation vector are classified as class 1
The confusion matrix format could be revised to make it easier to understand.
Thank you for formally accepting my answer.
Greg
  2 Comments
Michael Dorner
Michael Dorner on 14 Dec 2013
Edited: Michael Dorner on 14 Dec 2013
Thank you for answering. Unfortunately, I tried it with 50:50 data, but still the same result. The second (and further classes) are almost not detected, but they should, at least as wrong. In the sample code (e.g. iris_data) they are ok, even for two classes (although in this case the hitrate is quite bad :-) ).
Greg Heath
Greg Heath on 15 Dec 2013
Show your code, dimensions of input and target, comments and error messages

Sign in to comment.

More Answers (0)

Categories

Find more on Deep Learning Toolbox in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!