It is possible to train the data set of 100x15 in ANFIS

Dear all,

I am a very new user for ANFIS classifier.So, may I know it is possible to train 100x15 size of data set in .mat file. I trained this training data set ,and i occured error like " create very large rule based system" . May I know what 's the problem is? Thanks all.

Answer 1

[Krishna Chaitanya Duggineni]

This issue is a fundamental limitation and not a bug.

The following Web site provides information about the "curse of dimensionality":

http://www.cs.nthu.edu.tw/~jang/anfisfaq.htm

Question 5 in this page, titled "How come GENFIS1 generates a huge number of fuzzy rules? What is the 'curse of dimensionality?'" is relevant for this issue.

One way to reduce the number of rules generated for a large dataset is to use the GENFIS2 method could be like:

in = train_data(:,1:37);
out = train_data(:,38);
fismat = genfis2(in,out,0.8); % generate an initial FIS with 80 rules instead of 728
fis = anfis(train_data,fismat); % takes long time but avoids 'out of memory' error

Additionally, for large datasets, it is not recommend that we execute multiple trainings in a loop. To manage memory usage, I suggest using a separate script file for each training and executing "clear classes" at the MATLAB prompt between executing each of the scripts.

[Khaing Zin Htwe]

Thanks a lot ,sir. I have the result of training error 0.0002 and testing error is even 1.5 unfortunately.I used 55 inputs and 1 output datasets and genfis2. May I know how to use genfis2 efficiently and how can I reduce the testing error,please. If you have some resource links ,please give me,sir. THanks.