Are the following steps to calculate the accuracy correct or not?. Also, regarding to the sensitivity and specificity, should I do the same?
Mostra commenti meno recenti
I have a video clip that contains1500 frames. I trained my algorithm using a training video and detected the abnormalities in these frames. Thus, I have the detected frames which are logical frames and the ground truth dataset, which also is also logical frames. Now, I want to calculate the accuracy and ROC curve. I used the confusion function in this link (https://www.mathworks.com/matlabcentral/fileexchange/46035-confusionmatstats-group-grouphat-?focused=5540014&tab=function), which gives me the accuracy, sensitivity, and specificity. I applied the function to each row in the frame, then to each frame in the video to calculate the average accuracy.
6 Commenti
FAS
il 5 Feb 2018
FAS
il 5 Feb 2018
Rik
il 5 Feb 2018
It sounds like that depends on what you goal is. It doesn't sound logical to me to calculate acc, sens, and spec for each row, but it's your project, so you should decide.
As I understand ROC curves, they only work with a non-binary scale (either continuous or discrete), because you create it by increasing some threshold/metric and plotting the behavior of the sensitivity and specificity under those conditions. Your description doesn't sound like you have such a metric, so you can't produce an ROC curve.
And please don't use those comments with a few dots. They are only clutter, and the only effect they might have is annoy contributors, the same way urgent does.
FAS
il 6 Feb 2018
Rik
il 6 Feb 2018
Does it make sense to you in your case to calculate metrics for each row? if so, then there is nothing wrong with what you did.
Do you have some metric that has more than 2 possible values, so you need to find an optimal threshold to call something abnormal? If so, use that metric to make an ROC curve, if not, you can't make one.
Accuracy in the sense that we are using now has a very specific meaning that may or may not be useful for you to calculate. If this is a project where you can ask a supervisor for useful metrics, I advise you to do so.
FAS
il 6 Feb 2018
Risposte (0)
Categorie
Scopri di più su Detection in Centro assistenza e File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!