#5 ROC curve construction

[Scott Dick]

This RFE is for an advanced report called an ROC curve. An ROC curve is a plot of the False Positive rate against the True Positive rate for a binary classification dataset, given a variety of parameter settings for a machine learning algorithm. This is how performance is normally reported in machine learning research.

A point in ROC space is defined by the confusion matrix output after testing one trained classifier. Parameter settings for the classifier (such as the threshold for outputing class 0 or 1 from the GNN) are varied over a range, and a set of confusion matrices is generated. This yields a set of points in ROC space. These are then joined to form a piecewise linear curve, demonstrating how well the classifier solves the problem of minimizing false negatives and false positives.

There are two outcomes desired. In a GUI interactive session, we would like to see the actual ROC curve plotted on the screen. In a command-line session, the ROC curve data (individual points) should be an optional output.