scikit-learn Cookbook , Second Edition - Second Edition

Outline for future projects

Overall, for future classification projects, you can do the following:

1. Load the best data you can find for a particular problem.
2. Determine whether there is a classification context: is FP better or worse than FN?
3. Perform training and testing of the data without context using ROC-AUC scores. If several algorithms perform poorly at this step, you might want to go back to step 1.
4. If the context is important, explore it with confusion matrices.

Logistic regression is particularly well suited for all these steps, though any algorithm with a predict_proba() method will work very similarly. As an exercise, you can generalize this process for other algorithms or even general regression if you are ambitious. The main ...