Understanding Feature Engineering
Feature engineering is the step that leads to information extraction from existing data. It is extracted in terms of new features. For example, in the case of churn prediction of a telecommunication company the case "service" can contain different types of services (Internet service , DSL, Streaming TV, Streaming movies, etc.) so we can create a feature from every service type then use the one-hot-encoding method.
Some cases where feature transformation is required:
-Changing the scale of a variable from its original scale to scale between zero and one. We call this change a Normalization (Normalization : A normalized variable has a scale between 0 and 1)
-Since some algorithms give better results if we work with normally distributed data, we should deal with skewed datasets. There are different methods to do so like applying the log, square root or inverse of the values
Additional Resources for understanding feature engineering: