SVMs for multiclass classification

All of the same concerns that we had when we used SVC for binary classification apply when we are doing multiclass classification. We need to determine whether the classes are linearly separable, and if not, which kernel will yield the best results. As discussed in the first section of this chapter, we also need to decide whether that classification is best modeled as one-versus-one or one-versus-rest. One-versus-one finds decision boundaries that separate each class from each of the other classes. One-versus-rest finds decision boundaries that distinguish each class from all other instances. We try both approaches in this section.

We will work with the machine failure data that we worked with in previous chapters.

Note

This dataset on machine failure is available for public use at https://www.kaggle.com/datasets/shivamb/machine-predictive-maintenance-classification. There are 10,000 observations, 12 features, and two possible targets. One...