hope you enjoyed our previous newsletter of “Designing the Analytical Base Table” and how it helps in future of predictive data analytics. Let’s continue to our previous discussion with new topic “Designing and Implementing Features”.
2.4 Designing and Implementing Features
Once domain concepts have been agreed on, the next task is to design and implement concrete features based on these concepts.
What are Features?
Features are measurable data points we use to train machine learning models. They represent important ideas (domain concepts) from the real world.
For example, if the concept is "customer spending habits," a feature could be "average spending in the last 6 months."
Challenges in Creating Features:
Approximation: Sometimes, we can’t perfectly represent a Domain concept, so we use the best available data.
Proxy Features: If direct data isn’t available, we use related data instead (e.g., using customer reviews to estimate product satisfaction).
Concept Drop: If no data is available, we may need to skip that idea entirely.
Important Things to Consider:
Data Availability: Is the required data available? For example, if you need 6 months of account balances, the data must exist.
Timing: Can you get the data before making a prediction? For instance, attendance at a match can’t be used if you need predictions before the game starts.
Staying Useful Over Time: Will the feature remain relevant? Instead of using raw salary data (which changes over time), use a ratio like salary-to-loan amount.
Iterative Process:
Creating features isn’t a one-time job. We explore the data, design features, test them, and refine them repeatedly.
