100% precision in training set with SVM classifier

1 次查看（过去 30 天）
Jose Marques
Jose Marques 2018-3-12
评论： Jose Marques 2018-3-13
Hello! I have a Machine Learning doubt:
I am training a SVM classifier to classify binary images (70x70) in two classes. The dataset has 100 images/class.
In traning and test set the precision is 100%.
This makes me verify again the code and I make another experience: adding noise to the images e classifying again. With random images (100% noise) the precision in the training set was 100% e 50% approx.
The 100% training set precision with 100% noise is possible?
In the graph: Precision on test set
x-axis => noise;
y-axis => accuracy;
blue line => training the SVM classifier with noise
green line => training the SVM classifier without noise
  4 个评论
显示 2更早的评论
Walter Roberson
Walter Roberson 2018-3-12
You can get 100% accuracy in training of noise if you overfit. If your number of neurons is higher than the number of training datasets then it could potentially "remember" something characteristic of each data set.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Pattern Recognition and Classification 的更多信息

标签

Community Treasure Hunt

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

Start Hunting!

Translated by