How to improve Regression plot and error results for ANN in MATLAB?

2 views (last 30 days)
Hi All,
I am trying to train an ANN with 8 inputs and 2 outputs. I have tried to use one hidden layer with 30 neurons with TRAINLM as learning function. The problem I am getting is that although the R values of training, validation and test regression plots are greater than 0.98 (please see figure), the points are not concentrated on the linear line. Any ideas why this is happening. I have tried to use raw data as it is to train the ANN but that was not giving me good results. Then, I used Gaussian normalization for my inputs and outputs and I get the results that are shown in figures. The normalized error for both outputs are also attached.The normalized error ranges from 2 to -1.8.
Can you please guide how to solve this issue and improve my results? Am i giving too much data for training, testing and validation?
P.S. I am using MATLAB GUI for ANN by using nntool..
Thanks

Answers (0)

Community Treasure Hunt

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

Start Hunting!