In this chapter, we continued our journey of classifying news data with the SVM classifier, where we acquired the mechanics of an SVM, kernel techniques and implementations of SVM, and other important concepts of machine learning classification, including multiclass classification strategies and grid search, as well as useful tips for using an SVM (for example, choosing between kernels and tuning parameters). Then, we finally put into practice what we had learned in the form of two use cases: news topic classification and fetal state classification.

We have learned and adopted two classification algorithms so far, Naïve Bayes and SVM. Naïve Bayes is a simple algorithm (as its name implies). For a dataset with independent, or close to independent, features, Naïve Bayes will usually perform well. SVM is versatile and adaptive to the linear separability of data. In general, high accuracy can be achieved by SVM with the right kernel and parameters. However, this might be at the expense of intense computation and high memory consumption. When it comes to text classification, since text data is, in general, linearly separable, an SVM with linear kernels and Naïve Bayes often end up performing in a comparable way. In practice, we can simply try both and select the better one with optimal parameters.

In the next chapter, we will look at online advertising and predict whether a user will click through an ad. This will be accomplished by means of tree-based algorithms, including decision tree and random forest.