Beginning Anomaly Detection Using

IRU

FRO

LQ

QRUPDOFROXPQV

LI

QRUPDO>FRO@GW\SH   REMHFW

HQFRGHG /DEHO(QFRGHU

HQFRGHGILW QRUPDO>FRO@

QRUPDO>FRO@ HQFRGHGWUDQVIRUP QRUPDO>FRO@

IRU

FRO

LQ

QRYHOWLHVFROXPQV

LI

QRYHOWLHV>FRO@GW\SH  REMHFW

HQFRGHG /DEHO(QFRGHU

HQFRGHGILW QRYHOWLHV>FRO@

QRYHOWLHV>FRO@ HQFRGHGWUDQVIRUP QRYHOWLHV>FRO@

Figure 2-45.  Applying the label encoder to the data sets

IRU

I

LQ

UDQJH  

QRUPDO QRUPDOLORF>QSUDQGRPSHUPXWDWLRQ OHQ QRUPDO @

GI SGFRQFDW >QRUPDO>@QRUPDO>@@

GIBYDOLGDWH QRUPDO>@

[BWUDLQ[BWHVW WUDLQBWHVWBVSOLW GIWHVWBVL]H 

UDQGRPBVWDWH 

[BYDO GIBYDOLGDWH

Figure 2-46.  Shuffling the entries in the normal data set, and defining the 

training, testing, and validation sets

Chapter 2   traditional Methods of anoMaly deteCtion

---

66

You are only using a subset of the entire data set to train the model on because the 

larger the training data, the longer it takes for the OC-SVM to train.

Run the code in Figure 

2-48

 to declare and initialize the model.

By default, the 

kernel is set to ‘rbf’, meaning radial basis function. It is similar to the 

circular decision boundary that you saw in the earlier examples, and you use it here 

because you want to define a circular boundary around a set of regions that contain 

normal data. As seen in the earlier examples, any points that fall outside of the region 

are to be considered anomalies. 

Gamma tells the model how much you want to consider 

points further from the hyperplane. Since it is pretty small, this means you want to 

emphasize the points farther away. The