Beginning Anomaly Detection Using

confusion matrix (Figure 

2-1

). 

One thing to note is that in the case of anomaly detection, you only need a 2x2 confusion 

matrix since data points are either anomalies or they are normal data.

Chapter 2   traditional Methods of anoMaly deteCtion

---

27

From the values in each of the four squares, you can derive values for 

accuracy, 

precision, and recall to gain a better understanding of how your model performs.

Here’s the confusion matrix with all the formulas (Figure 

2-2

):

Figure 2-1.  Confusion matrix

Chapter 2   traditional Methods of anoMaly deteCtion

---

28

• 

Precision is a measurement that describes how many of your true 

predictions actually turned out to be true. In other words, for all of 

your true predictions, how many did the model get right?

• 

Accuracy is a measurement that describes how many predictions 

you got right over the entire data set. In other words, for the entire 

data set, how many did the model correctly predict were positive  

and negative?

• 

Recall is a measurement that describes how many you predicted true 

for all data points that were actually true. In other words, for all of 

the true data points in the data set, how many of them did the model 

predict correctly?

Figure 2-2.  Precision, Accuracy and Recall

Chapter 2   traditional Methods of anoMaly deteCtion

---

29

From here, you can derive more values.

F1 Score is the harmonic mean of precision and recall. It’s a metric that can tell us 

how accurate the model is, since it takes into account both how well the model makes 

true predictions that are actually true, and how many of the total true predictions that 

the model correctly predicted.

 

F Score

1

2

=

*

*

+

Precision Recall

Precision Recall

 

The