Feature Engineering Intuition

Feature engineering defined

Feature engineering is the process of using business domain and technical knowledge to extract features from raw data. Extracted features transform raw data into a format that can improve the predictive performance of machine learning algorithms.

This definition has essential implications for machine learning practitioners:

• Machine learning algorithms are not guaranteed to learn from raw data.

• Machine learning algorithms might learn the wrong things from raw data.

• Feature engineering is an art where domain expertise and technical knowledge are vital.

• Feature engineering is required to craft the most valuable machine learning models.

Feature engineering example

When analyzing data, especially business data, using date / time data (e.g., timestamps) is very common. Examples include timestamps for order creation, customer service calls, and website visits. Timestamps represent potent pieces of information that machine learning models can use.

However, using raw timestamps to train machine learning models usually leads to models that overfit. The following table is an example of how timestamps are commonly represented.