DataEval: An End-to-End Data Analysis Tool for Machine Learning Development
Saturday, April 18, 2026 · 3:30 PM – 5:00 PM · Classroom 205
Data is fundamental to learning in ML/AI tasks. At its core, the motivation to evaluate your dataset is simple: if the training data is flawed, it’s likely the resulting model will be inherently unreliable. While a common sentiment suggests that “more data” is a universal fix for poor performance, many operational environments lack resources to accommodate massive datasets. This raises a critical question of how can you determine whether your dataset is “good enough” for a specific use case before committing significant time and resources to the training process? Beyond resource constraints, another challenge lies in simply knowing how to evaluate dataset quality. There is a vast number of statistical tests available and the relevance of each varies based on data types and model architectures. Without a clear, industry-wide standard for evaluation practices, scientists and engineers often struggle to identify the most effective starting point for data verification and validation.
DataEval bridges this gap by providing a suite of tools that quantify data quality at each stage of the ML lifecycle. It organizes dataset assessment into broad categories such as quality, distribution, coverage, and completeness. DataEval supports informed decisions about dataset adequacy, helping determine whether the dataset is suitable for the intended operational environment and guiding targeted data collection or relabeling when gaps are identified. Furthermore, DataEval provides users with guidance and documentation on how to navigate each step of the process from data cleaning, to model development, to monitoring deployed systems.
What is DataEval? DataEval is an open-source Python-based library developed under the Joint AI Test Infrastructure Capability (JATIC) program and provides metrics that support the verification and validation of data for developing models.
Who is DataEval for? We aim to demonstrate the tool at a high level, which will be accessible to attendees at the introductory/intermediate level. However, the core functions of DataEval are capable of supporting even advanced users.
By the end of this session, attendees will (1) understand the critical role of data evaluation in the ML lifecycle, (2) be able to navigate the DataEval library, and (2) have the skills to execute a streamlined data cleaning and evaluation workflow on a real-world dataset.
Resources: https://drive.google.com/drive/folders/13JjUX65MCAC0GA1GNJqIUQifMsmlf0IZ?usp=drive_link
About the Speaker
Haley Green
ARiA, Research & Design Scientist
Dr. Haley Green is a Research & Design Scientist as ARiA. She is a graduate of the University of Virginia, where she earned her Ph.D. in Computer Engineering. Prior to UVA, Green earned her B.S. in Mechanical Engineering at Brown University while playing for the Women’s Basketball team. Green’s research background is in human-robot interaction, and she’s currently working on AI assurance at ARiA. She has organized workshops for Robotics: Science and Systems and presented at AAAI, HRI, and ICRA.