Measuring Model Accuracy

Quality of classification model predictions

So far, this course has relied on accuracy to determine the value of CART classification tree models. Using accuracy has been helpful because a large audience easily understands the accuracy measure (i.e., the percentage of correct predictions).

However, despite accuracy being the default metric for classification problems, there are other, or better, metrics to evaluate the quality of predictions. This lesson covers several ways of measuring the quality of classification model predictions:

• Accuracy

• Confusion matrices

• Sensitivity

• Specificity

Note: While this lesson uses binary classification examples, the techniques also apply to multiclass scenarios (i.e., three or more unique label values).

Accuracy

The most intuitive measure of classification model predictions is accuracy. Simple put, accuracy is the percentage of labels predicted correctly: