I am using matlab R2017a, and I am having a dataset of 12 videos. I have plotted the ROC curve using perfcurve command as below:
[X,Y,~,AUC] = perfcurve(labels,scores,true);
But when I plotted it, the output was so disturbing in this form. It was not according to the requirement.
I have searched solutions to this problem. A solution I found states that dataset is too little. But I cannot increase my data as I have to test it on 12 videos.
Kindly guide me what should I do then?
It will be a great favor
Hi colleague
Kindly, I supposed your data are very large, but unfortunately I saw too little.
Regards....
Mohammed Deeb My train data contain 16 video and test data contain 12 videos. The content of "Required_ROC.png" is removed because it contains two more graphs, but I have to consider only this graph which is shown. The data is processed using optical flow method, HOF descriptor, and markov model. As you are saying that plot seems normal and acceptable, then is there any way to smooth the output of "wrong_ROC.png" instead of increasing the input data
Mohammed Deeb Thanks for your suggestion. I have to re-implement the output of a research paper. That's why I cant use CNN. I have code of that research paper, I have implemented it in the same way as described by them. But unfortunately, my output ROC curve is not same as theirs.
- Badges
- Science topic