i have dataset which contains data from 10 subject. My idea fir cross validaytion is leave one person out cross validation. Here i trian on data from 9 subjects and test on data from 1. When we normally do cross validation, we have a stopping criteria which avoids model overfitting.
How do I avoid overfittiing in my case.
below is code snippet
for idx = 1:N%k = LOOCV train on rest; validate on K- meal
s = [1:idx-1 idx+1:N];
Xtrain= Training(s); %(all remaining datasets)
Xvalidate = Training(idx);% idx dataset
Xtrainlabel = Training_labels(s);
Xvalidatelabel = Training_labels(idx);
Mdl = fitcsvm(XTrain(:,featsel),...
XTrainlabel);
[trainSVM,trainScoreSVM] = resubPredict(Mdl); %training
%- Cross-validate the classifier
CVSVMModel = crossval( Mdl );
%validation
Yval_pred= predict(Mdl, XValidate(:, featsel)); %validation
[cmV,order] = confusionmat(Yval_pred, actual_val);
tnV = cmV(1,1);
fnV = cmV(1,2);
fpV = cmV(2,1);
tpV = cmV(2,2);
Accuracy(idx) = (tp+fp)./(tp+fp+tn+fn);
end

2 commentaires

Wan Ji
Wan Ji le 20 Août 2021
Use dropoutLayer may help you avoid model overfitting. Try it
pallavi patil
pallavi patil le 20 Août 2021
i am using svm as classifier. I supoose dropoutLayer works for neural network.

Connectez-vous pour commenter.

Réponses (1)

Prince Kumar
Prince Kumar le 7 Sep 2021

0 votes

You can try the following methods:
  1. Remove features
  2. Feature Selection
  3. Regularization
  4. Ensemble models if you are ok with trying models other than SVM

Catégories

En savoir plus sur Deep Learning Toolbox dans Centre d'aide et File Exchange

Produits

Version

R2020b

Community Treasure Hunt

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

Start Hunting!

Translated by