Follow the Herd: A Data-Driven Approach for Actionable Recourse
What it is
This paper proposes Data-Driven Action Recourse (DDAR), a method for generating actionable recourse for machine learning decisions without requiring hand-designed cost functions. It builds a dynamics model from historical user trajectories, assuming that highly likely actions are lower cost, and searches for low-cost paths to desired outcomes.
The problem it solves
Existing recourse methods rely on expensive human supervision to hand-design cost functions, which is costly, subjective, and difficult to scale.
What you could use it for
- Providing actionable advice to users rejected by credit or insurance models
- Generating low-cost action plans for users to improve their ML model scores
- Automating recourse generation in high-stakes decision systems
Why it matters
Reduces the cost and bias of generating recourse by eliminating the need for expensive human supervision to define action costs, potentially making ML recourse more scalable and equitable.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.