Fair Active Learning: Solving the Labeling Problem in Insurance
Romuald Elie, Caroline Hillairet, François Hu, Marc Juillard
TL;DR
The paper tackles the labeling bottleneck and fairness challenges in insurance ML by integrating active learning with a fairness constraint, focusing on Demographic Parity to reduce bias. It introduces a model-agnostic fair active learning (FAL) approach that derives fair predictions and uses them to guide informative sampling, achieving a favorable balance between accuracy and fairness on both synthetic and actuarial datasets. The work analyzes classic AL methods, demonstrates their limitations for fairness, and provides theoretical and empirical results showing how FAL can mitigate bias while preserving predictive performance. This approach has practical implications for actuarial practice and AI governance, offering a scalable path to fair, data-efficient pricing and risk assessment tools.
Abstract
This paper addresses significant obstacles that arise from the widespread use of machine learning models in the insurance industry, with a specific focus on promoting fairness. The initial challenge lies in effectively leveraging unlabeled data in insurance while reducing the labeling effort and emphasizing data relevance through active learning techniques. The paper explores various active learning sampling methodologies and evaluates their impact on both synthetic and real insurance datasets. This analysis highlights the difficulty of achieving fair model inferences, as machine learning models may replicate biases and discrimination found in the underlying data. To tackle these interconnected challenges, the paper introduces an innovative fair active learning method. The proposed approach samples informative and fair instances, achieving a good balance between model predictive performance and fairness, as confirmed by numerical experiments on insurance datasets.