Quality Control
As a quality measure of the implemented feature extraction methods, the commonly used in statistics ROC and DET curves are implemented. Ground truth data are expected to be provided by the user.
ROC Curves
Receiver Operating Characteristics
Probability curves, plotting the True Positive (TP) rate against to the false positive (FP). True Positives (y-axis) is defined as the percentage of correct matches over the total number of matches while the False positive (x-axis) is the respective percentage of incorrect matches.
The highest the AUC (Area Under the Curve) index is, the better our results are.
True Positive Rate = TP/(TP+FN)
For more details see Google Developers
DET Curves
Detection Error Trade-off
Commonly used in classificatio problems, it plots the false negative rate against the false positive rate. The lines shows the trade-off between the above.
Written with StackEdit.