In this chapter, we discussed supervised machine learning approaches to solving classification tasks. These approaches use trained models to determine the class of an object according to its characteristics. We considered two methods of binary classification: logistic regression and SVMs. We looked at the approaches for the implementation of multi-class classification with the use of binary classifiers.

We also examined the nearest neighbor method, which can deal with multi-class classification without additional actions. We saw that working with non-linear data requires additional improvements in the algorithms and their tuning. Implementations of classification algorithms differ in terms of performance, as well as the amount of required memory and the amount of time required for learning. Therefore, the classification algorithm's choice should be guided by a specific task and business requirements. Furthermore, their implementations in different libraries can produce different results, even for the same algorithm. Therefore, it makes sense to have several libraries for your software.

In the next chapter, we will discuss recommender systems. We will see how they work, which algorithms exist for their implementation, and how to train and evaluate them. In the simplest sense, recommender systems are used to predict which objects (goods or services) are of interest to a user. Examples of such systems can be seen in many online stores such as Amazon or on streaming sites such as Netflix, which recommend you new content based on your previous consumption.