Problem Framing

Problem Framing is one of the most critical processes in any ML project. It bridges the gap between what the business needs and what ML can deliver. A poorly framed problem leads to wasted resources, missed expectations, and failed projects.

This process involves translating business requirements into a precise ML problem statement, selecting the appropriate modeling approach, defining inputs and outputs, and establishing clear boundaries for what the solution will and won't do.

Understanding the type of ML problem you're solving is fundamental. Each type requires different algorithms, data structures, and evaluation approaches:

• 1
  Business-to-ML Translation
  Convert business objectives into specific ML tasks. Identify what needs to be predicted, classified, or generated, and how it connects to business value.
• 2
  Problem Type Selection
  Determine the appropriate ML problem type (classification, regression, clustering, etc.) based on the desired output and available data characteristics.
• 3
  Input/Output Definition
  Clearly specify what features (inputs) will be used and what the model should produce (outputs). Define data types, formats, and expected ranges.
• 4
  Scope Delimitation
  Establish clear boundaries: what the model will and won't do, which use cases are in scope, and what edge cases should be handled.
• 5
  Hypothesis Formulation
  Document key assumptions and hypotheses about the problem. What patterns do you expect to find? What relationships should exist in the data?
• 6
  Constraint Identification
  Identify technical, business, and ethical constraints. Consider latency requirements, explainability needs, fairness criteria, and regulatory compliance.

Use this structured approach to systematically frame your ML problem:

• ✓
  Start Simple
  Begin with the simplest version of the problem. You can always add complexity later. A working simple model beats a complex model that never ships.
• ✓
  Challenge the Problem
  Question whether ML is really needed. Sometimes rule-based systems or simpler analytics can solve the problem more efficiently.
• ✓
  Consider the Human Baseline
  Understand how humans currently solve this problem. This provides context for what ML needs to beat and reveals important domain knowledge.
• ✓
  Think About Edge Cases Early
  Identify edge cases and exceptions during framing. They often reveal gaps in understanding and prevent surprises during development.
• ✓
  Document Assumptions Explicitly
  Every assumption is a potential point of failure. Document them clearly so they can be validated and revisited.