Beginning Anomaly Detection Using

bad margin or suboptimal margin has 

the hyperplane too close to one class or the distance not be as far as it can be to the 

hyperplane for each point or support vector.

As for the one-class support vector machine, Figure 

2-40

 shows what the graph 

would look like.

During training, the OC-SVM learns the decision boundary for normal observations, 

accounting for a few outliers. If 

novelties, new data points that the model has never seen 

before, fall within this decision boundary, they are considered normal by the model. If 

they fall outside of the boundary, they are considered anomalous. This technique is an 

example of semi-supervised novelty detection, where the goal is to train the model on 

normal data, and then it attempts to find anomalies in new data.

By doing so, the OC-SVM can capture the shape of the data pretty well thanks to the 

decision boundary that captures most of the training observations.

Figure 2-40.  An example of the decision boundary for a one-class support vector 

machine

Chapter 2   traditional Methods of anoMaly deteCtion

---

63