Comparative summary

There are a large number of models and algorithms for supervised statistical learning, and this article has limited itself to presenting the main ones. As mentioned, each has its own advantages and disadvantages. The choice of one technique over another for a given application is therefore always a delicate one. Ideally, at least the main ones should be tried and compared, but this can be a major undertaking. All the more so as you need to be sure of finding the optimum values for each of the sometimes numerous hyper-parameters, using a validation base (or cross-validation).

However, the characteristics of the data to be processed can guide the choice of model (figure 32 ):