Here I am again. You just took another major leap on your journey to machine learning discovery. If you took the right time to acquire and practice what Andy just showed you, you should have now added to your toolbox two of the most employed classification models: logistic regression and support vector machines. Both of them are employed to perform classification exercises.

The logistic regression predicts the probability of a given outcome occurring, estimating the level of contribution to this output provided by all of the explanatory variables. This makes this model quite useful when interpretability is one of the objectives of the analysis.

On the other side, you have support vector machines, which are based on the concept of a hyperplane, a sort of blade of different possible shapes able to divide our population into two or more groups, and by that, mean perform the desired classification task. This algorithm shows pretty high performance, especially with a non-linear hyperplane, but on the other side also shows a lower level of interpretability.

For both models, you have also learned what the relevant assumptions are and how to test them. Among these assumptions, one that should be considered as really relevant is the one about the independent and identically distributed variables, which is considered a requisite for the vast majority of machine learning methods.

Finally, you have acquired new elements to solve our mystery, since both of the models have shown acceptable performance and significance, and could be employed to derive the final list of the probability of defaulted companies to focus subsequent internal audit analyses on.