Match point! You and Andy finally got the list Mr Clough requested. As you may be guessing from the simple fact that there are still pages left in the book, this is not the end.

All you know at the moment is that the companies that probably produced that dramatic drop in Hippalus revenues are small companies with previous experiences of default and bad ROS values. We could infer that those are not exactly the ideal customers for a wholesale company such as Hippalus. Why is the company so exposed to these kinds of counterparts?

We actually don't know at the moment: our data mining models got us to the entrance of the crime scene and left us there. What would you do next? You can bet Mr Clough is not going to let things remain that unclear, so let's see what happens in a few pages.

In the meantime, I would like to recap what you have learned in this chapter. After learning what decision trees are and what their main limitations are, you discovered what a random forest is and how it overcomes those limitations. You also learned how to apply this model to your data with R, through the randomForest() function within the randomForest package.

Employing the objects resulting from svm(), glm(), and the randomForest() package, you learned how to obtain a confusion matrix and a lot of related metrics such as accuracy and precision. To do so, you employed the confusionMatrix() function contained within the caret package.

You also learned what ensemble learning is and why it is so useful. Applying it to the estimated classification models, you discovered how effective it can be in improving your final performance.

Finally, you learned how to perform predictions on new data by employing trained models, through the predict.something function.