Beginning Anomaly Detection Using

Figure 2-22.  Getting the anomaly scores from the trained isolation forest model 

and plotting a histogram

Figure 2-23.  A histogram plotting the average path lengths for the data points.  

It helps you determine what is an anomaly by using the shortest set of path lengths, 

since that indicates that the model was able to easily isolate those points

Chapter 2   traditional Methods of anoMaly deteCtion

---

49

A quick note: plt.show() is not necessary on Jupyter if you have %matplotlib inline, 

but if you are using anything else, this should open up a new window with the graph.

Let’s calculate the 

AUC to see how well the model did. Looking at the graph, there 

appears to be a few anomalous data with average path of less than -0.15. You expect 

there to be a few outliers within the normal range of data, so let’s pick something more 

extreme, such as -0.19. Remember that the lesser the path length, the more likely the 

data is to be anomalous, hence why there’s a curve that increases drastically as the graph 

goes right. Run the code in Figure 

2-24

.

You should see something like Figure 

2-25

.

That’s an impressive score! But could it be the result of overfitting? Let’s get the 

anomaly scores of the test set to find out.

Run the code in Figure 

2-26

.