Appendix B. K-nearest neighbors and support vector machines
In this appendix, we will examine classical machine learning algorithms with a more computational nature that we didn’t treat in the book because they are less frequently used nowadays and considered outdated compared to decision tree ensembles in most applications, but a few ones. Overall, SVMs are still a practical machine learning algorithm well-suited for high-dimensional, noisy, or small-sized data applications. On the other end, k-NN is well-suited for running applications where the data has few features, there can be outliers, and it is unnecessary to get a high degree of accuracy in predictions. For instance, SVMs can still be used to classify medical images, such as mammograms and X-rays, for vehicle detection and tracking in the automotive industry or to detect email spam. Instead, k-NN is mainly applied in recommender systems, particularly collaborative filtering approaches, to recommend products or services based on users' past behavior.