How to do when the inputs ranges for neural network are not so uniform in magnitude ?

Hi,i'm trying to create neural network for classification of EEG signal in which i extacted 9 features that i will use it as inputs for neural network but the problem is that my inputs ranges are not so uniform in magnitude i mean some input can take a very high value numerically compared to other inputs so the big numbers will mask the effect of the smaller numbers. I used the preprocessing function 'mapminmax' but i still have a bad performance so when i do some research i found that i should Transform each input that has always a positive value into logarithm values. Matlab will take these transformed values and apply the normalization to them [-1,1]. For prediction, i should enter not the actual values but the logarithm of the input to the ANN. So i want tp know is this right ? and can i do this transformation of my innputs with matlab?
This how my original input looks like

 Accepted Answer

Typically, MAPMINMAX only fails when outliers are present.
Try MAPSTD to identify and modify (or delete?) outliers
Hope this helps.
Thank you for formally accepting my answer
Greg

9 Comments

Did you mean MAPSTD??? because when i search for MAPST it i coldn't find it
i tried to use MAPSTD without logarithme transformation and with logarithme transformation and i found that the accuracy improve just little bit with logarithme transformation but not still very good . Is it logical?
When you used mapstd on the log values what were the 9 minimum and maximum ranges?
i don't understand your question ver well but i think that the ranges of all the values will be [-1,1] isn't?
No.
MAPSTD yields zero mean and unit variance. Mins and Maxes will be different. I use this for outlier detection.
MAPMINMAX yields [ -1 1 ] mins and maxes. Means and Variances will be different.
Hope this helps.
Greg
this is an example of log transformed data
and this is after application of MAPSTD
i want to know is it correct to use it for my case ?
I can't tell with tabulations. I tend to relay on linear and scatter plots.
Greg
How can i do this linear and scatter plots ?

Sign in to comment.

More Answers (0)

Categories

Asked:

on 14 Jun 2017

Commented:

on 28 Jun 2017

Community Treasure Hunt

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

Start Hunting!