Beginning Anomaly Detection Using

kernel trick. There are different types of kernels, including the linear 

kernel you saw in the earlier examples. Other types of kernels include polynomial 

kernels, which map the data to some nth dimension using a polynomial function, and 

exponential kernels, which map the data according to an exponential function.

Another term to cover is 

regularization, a parameter that tells the SVM how much 

you want to avoid misclassifications. 

Lower regularization values lead to graphs like 

the one you saw earlier where there were a few outliers on either side of the hyperplane. 

Higher regularization values lead to graphs where you saw the hyperplane separate 

every single point, at the cost of possibly overfitting on the data.

Gamma tells the SVM how much to consider points farther away from the region 

of separation between the classes. 

Higher gamma values tell the SVM to only consider 

nearby points, while 

lower gamma values tell the SVM to also consider the points 

farther away.

Figure 2-39.  This is what the hyperplane looks like when you go back to 2D

Chapter 2   traditional Methods of anoMaly deteCtion

---

62

Finally, the 

margin is the separation between each class and the hyperplane. As 

discussed earlier, an 

ideal margin involves the maximum equidistant separation of 

each of the closest from the hyperplane. A