Applying feature engineering

In real-world projects, what can make the difference between a successful machine learning model and a mediocre one is often the data, not the model. When we talk about data, the differentiator between bad, good, and excellent data is not just the lack of missing values and the reliability of the values (its “quality”), or the number of available examples (its “quantity”). In our experience, the real differentiator is the informational value of the content itself, which is represented by the type of features.

The features are the real clay to mold in a data science project, because they contain the information that models use to separate the classes or estimate the values. Every model has an expressiveness and an ability to transform features into predictions, but if you are lacking on the side of features, no model can bootstrap you and offer better predictions. Models only make apparent the value in data. They are not...