Integration of Data Quality Metrics with Machine Learning Models for Robust Decision-Making Systems

Abstract

The efficacy of Machine Learning (ML) models is fundamentally constrained by the quality of the data upon which they are trained and deployed. This paper explores the critical integration of systematic Data Quality (DQ) metrics into the ML pipeline to build robust decision-making systems. We argue that moving beyond ad-hoc data cleaning to a continuous, metrics-driven assessment of data across its lifecycle—encompassing dimensions such as accuracy, completeness, consistency, and timeliness—is essential for improving model performance, fairness, and reliability. We present a framework for embedding DQ checks at key stages: pre-modeling (data profiling), in-training (monitoring for drift and anomalies), and post-deployment (feedback loops). Through conceptual diagrams and analysis, we demonstrate how quantified DQ scores can be used to trigger automated remediation, weight training instances, enrich feature sets, and provide interpretable diagnostics for model predictions. This integration fosters greater trust in AI systems by making their dependence on data quality explicit and manageable.