Adaptive method improves finding wrong labels in tabular data
Adaptive Ensemble Selection for Noisy Labels on Tabular Data
Machine Learning
Summary
Mistakes in data labels can make learning from data less accurate, especially when these mistakes are hard to spot. The authors suggest a system that smartly combines several ways to detect these errors, choosing the best methods based on the data itself. Their approach works about as well as a strong existing method and sometimes does better depending on the data. They also found that how well each detection method works depends on specific details of the data. This shows that automatically picking detection tools based on data properties can help create more trustworthy data pipelines.
What this means in practice
- •For data science teams: Improve reliability of supervised machine learning by selecting tailored noise detection strategies for tabular datasets with corrupted labels.
- •For healthcare data teams: Enhance quality checks on medical tabular data where label errors can be subtle and harm predictive model accuracy.
Authors
Faizaan Ali, Inwon Kang, Oshani Seneviratne
Abstract
Incorrect or corrupted labels in tabular datasets can significantly degrade supervised learning performance, particularly when mislabeling is subtle and not easily detectable from feature space alone. In the context of automated or AI-augmented data science workflows, robust detection of such label noise is critical for building reliable models. We propose a data-centric reasoning module for AI data science systems that automatically diagnoses dataset quality and selects appropriate cleaning strategies. Given a dataset, a meta-model predicts weights over a diverse set of detectors, including confidence-based, neighborhood-based, and distributional methods. Across benchmark datasets with controlled noise, our approach achieves performance comparable to a Confident Learning baseline on average, with dataset-dependent gains and losses, particularly in heterogeneous regimes. We further show that detector effectiveness is systematically linked to dataset properties. These results demonstrate the value of descriptor-driven, data-centric ensembling as a component of AI-assisted data-science pipelines for robust dataset assessment and model reliability.