Abstract

This paper addresses the critical role of automated data quality monitoring in Machine Learning Operations (MLOps) pipelines. As organizations increasingly rely on machine learning models for decision-making, ensuring the quality and reliability of input data becomes paramount. The paper explores various types of data quality issues, including missing values, outliers, data drift, and integrity violations, and their potential impact on model performance. It then examines automated detection methods, such as statistical analysis, machine learning-based anomaly detection, rule-based systems, and data profiling. The integration of data quality monitoring into different stages of the MLOps pipeline is discussed, emphasizing continuous monitoring at data ingestion, pre-training validation, post- deployment drift detection, and feedback loops for model retraining. The paper also addresses key challenges in implementing automated data quality monitoring, including balancing precision and recall in anomaly detection, handling high-dimensional and unstructured data, managing false positives and alert fatigue, and adapting to evolving data distributions. By providing a comprehensive framework for automating data quality monitoring in MLOps pipelines, this paper aims to equip practitioners with the knowledge and strategies necessary to enhance the reliability and performance of machine learning systems in production environments.