Predictive Multiplicity in Classification
arXiv:1909.06677
Abstract
Prediction problems often admit competing models that perform almost equally well. This effect challenges key assumptions in machine learning when competing models assign conflicting predictions. In this paper, we define predictive multiplicity as the ability of a prediction problem to admit competing models with conflicting predictions. We introduce formal measures to evaluate the severity of predictive multiplicity and develop integer programming tools to compute them exactly for linear classification problems. We apply our tools to measure predictive multiplicity in recidivism prediction problems. Our results show that real-world datasets may admit competing models that assign wildly conflicting predictions, and motivate the need to measure and report predictive multiplicity in model development.
References in corpus (10)
- All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously
- Model Selection Techniques -- An Overview
- Supersparse Linear Integer Models for Optimized Medical Scoring Systems
- The Dataset Nutrition Label: A Framework To Drive Higher Data Quality Standards
- Does mitigating ML's impact disparity require treatment disparity?
- Interventions over Predictions: Reframing the Ethical Debate for Actuarial Risk Assessment
- Aequitas: A Bias and Fairness Audit Toolkit
- Fairwashing: the risk of rationalization
- Variable Importance Clouds: A Way to Explore Variable Importance for the Set of Good Models
- Disentangling Influence: Using Disentangled Representations to Audit Model Predictions
Cited by in corpus (11)
- Underspecification Presents Challenges for Credibility in Modern Machine Learning
- On the Existence of Simpler Machine Learning Models
- Visual Analysis of Discrimination in Machine Learning
- Directional Bias Amplification
- On Counterfactual Explanations under Predictive Multiplicity
- Characterizing Fairness Over the Set of Good Models Under Selective Labels
- Improvement-Focused Causal Recourse (ICR)
- Consistent Counterfactuals for Deep Models
- DIVINE: Diverse Influential Training Points for Data Visualization and Model Refinement
- Selective Ensembles for Consistent Predictions
- A Series of Unfortunate Counterfactual Events: the Role of Time in Counterfactual Explanations