How do I evaluate the performance of Haar Cascade in test data ? How do we classify the set of images in different categories using Haar Cascade, just like CNN does?

edit

How do we classify the set of images in different categories like cat and dog using Haar Cascade, just like CNN does?

you can't do this.

cascade classifiers are binary, detect a single class, or the absence of it.

if you want to build a confusion matrix for a single class(ifier), you'll need annotated images (ground truth face boxes), test those with your trained classifier, and evaluate the detections using IOU

long story short: IF you have several classes to detect, use a cnn, not cascades.