The Effects of Data Quality on Machine Learning Performance on Tabular Data

Data quality directly impacts AI reliability: The paper demonstrates that incomplete, erroneous, or inappropriate training data leads to unreliable models and poor decisions, highlighting that trustworthy AI requires high-quality data across multiple dimensions including accuracy, completeness, and consistency.

Different types of data pollution have distinct effects: The research distinguishes between three scenarios—polluted training data, polluted test data, or both—suggesting that where quality problems occur in the AI pipeline matters significantly for model performance, with implications for where organizations should focus quality control efforts.

Quality issues affect multiple algorithm types: By testing 19 popular machine learning algorithms across classification, regression, and clustering tasks, the paper shows that data quality problems are not isolated concerns for specific AI applications but represent a broad challenge across different types of AI systems.

Six dimensions of data quality matter: The study examines data quality across six dimensions (accuracy, completeness, consistency, and others), indicating that responsible AI requires attention to multiple aspects of data quality rather than focusing on just one or two factors.

Data quality is foundational to responsible AI: The paper emphasizes that trustworthy AI applications cannot be achieved through algorithm improvements alone—high-quality training and test data along multiple dimensions is a prerequisite for responsible AI deployment.