Problem Framing: From Business Need to ML Task
What you'll learn: How to translate a real-world business problem into a well-defined machine learning task.
The Translation Challenge
You've learned what machine learning is and the different learning paradigms that exist. But before you can apply any ML technique, you must answer a critical question: "What exactly am I asking the model to do?"
Problem framing is the art of converting a vague business need like "increase customer satisfaction" or "reduce costs" into a concrete ML task with clear inputs, outputs, and success metrics.
The Core Translation Process
Think of problem framing like translating spoken language into a specific recipe:
- Identify the business goal — What outcome matters? (e.g., "Reduce customer churn")
- Define what success looks like — How will you measure it? (e.g., "Predict which customers will leave next month")
- Choose the ML task type — Based on your prior knowledge:
- Supervised learning if you have labeled examples (classification: will churn yes/no?)
- Unsupervised learning if you need to find patterns (clustering: discover customer segments)
- Reinforcement learning if decisions happen over time (optimize email timing)
- Specify inputs and outputs clearly — What features describe each customer? What exactly should the model predict?
Common Pitfalls
- Framing too broadly: "Make everything better" isn't actionable
- Choosing ML when a simple rule would work (violates "When to Use ML vs Traditional Programming")
- Forgetting that the model is part of a larger system (recall: ML in Context)
Key Takeaway: Problem framing transforms ambiguous business needs into precise ML tasks by defining measurable objectives, selecting appropriate learning paradigms, and clearly specifying inputs and outputs—it's the essential first step that determines whether your ML solution will succeed.