Classifier Measures

Accuracy

We can define the accuracy we’ve measured thus far as:

$$Accuracy = \frac{\sum TruePositives + \sum TrueNegative}{TotalPopulation}$$

It does not care whether there is a systematic error, such as the algorithm’s inability to correctly predict a passenger who survived (true positives), as long as it performs well at correctly predicting passengers who die (true negatives).

The confusion matrix offers us more detailed measures of our classifier performance.

These are:

• Precision
• Recall
• Specificity
• Negative predictive value (NPV)