ReClean: Reinforcement Learning for Automated Data Cleaning in ML Pipelines*

Addressing data quality issues is a challenging task due to the labor-intensive nature of manual data cleaning processes and the inadequacy of automated tools that lack effective repair strategies. In this paper, we introduce ReClean, a novel automated data-cleaning method, dedicated to ML pipelines, that employs reinforcement learning (RL) to optimize data-cleaning tasks. ReClean treats data cleaning as a sequential decision process, where RL agents learn to choose optimal data repair operations that improve ML model convergence and predictive performance. Our extensive experimental evaluation shows that ReClean surpasses existing baseline methods, successfully determining and applying data repair tools to enhance downstream predictive tasks automatically and without supervision.