How can i calculate sensitivity(True positive rate) and specificity(True negative rate) using libsvm as a binary classifier in matlab? -

i working on face verification problem.i using libsvm classifier. want calculate true positive rate , true negative rate.

by using these 2 performance measures, want calculate equal error rate , want draw roc curve.

i read perfcurve command in matlab.but here score means in command.??

if using libsvm, there 3 possible return values svmpredict function.

[predicted_label, accuracy, decision_values] = svmpredict(testing_label_vector, testing_instance_matrix, model [,'libsvm_options']);

if don't specify want return values, assigning output using several variables, first variable, predicted_label.

if want produce roc curve, need classifier scores each instance in order calculate thresholds. scores decision_values.

you can use either roc or plotroc neural network toolkit or perfcurve machine learning toolkit generate roc curve.

for true positive , false positive rates @ each threshold in form of cell array, use

[tpr,fpr,thresholds] = roc(ground_truth, decision_values);

for plot of roc curve, use

plotroc(ground_truth, decision_values);

[x,y] = perfcurve(ground_truth, decision_values, positive_class_name); plot(x,y);

Braziel