CA-MOO-Clean: Constraint-Aware Multi-Objective Optimization for Data Cleaning

Abstract

Automated data cleaning systems face an inherent tension: maximizing repair accuracy often requires aggressive modifications that violate the minimum change principle, while preserving data integrity risks missing critical errors. Existing methods optimize a single objective, failing to capture the multi-dimensional nature of this trade-off. We propose CA-MOO-Clean, a framework that formalizes data cleaning as a Constrained Multi-Objective Optimization problem. Our approach integrates NSGA-II evolutionary search with LLM-assisted constraint discovery to automatically generate Pareto-optimal cleaning pipelines, revealing the complete accuracy-modification trade-off frontier. Extensive experiments demonstrate that CA-MOO-Clean achieves competitive or superior accuracy while providing multiple Pareto-optimal solutions per run—93.85% Repair F1 on Hospital (5.91% modification), 86.73% on Adult (0.95% modification), and 93.92% on Credit (2.62% modification). Unlike single-solution baselines (HoloClean, NADEEF, AlphaClean), our framework reveals the full accuracy–modification trade-off frontier, achieving the highest accuracy on all three datasets while providing flexible trade-off selection without re-optimization. The architectural separation of one-time discovery (7.0–68.1s) from rapid execution (0.30–50.20s) enables efficient amortized processing. This work establishes multi-objective evolutionary optimization as a principled methodology for balancing accuracy and data preservation in automated cleaning systems.

Keywords

Data cleaning / Multi-objective optimization / Large language models / Automated data pipelines / Data quality constraints

RIGHTS & PERMISSIONS

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.