Rostami receives $288K USDA NIFA grant for study on food authenticity

Tuesday, Nov 10, 2026 • Greg Pederson :

A data scientist at The University of Texas at Arlington has received federal funding for a project which incorporates explainable artificial intelligence into agricultural data science, with the long-term goal of helping consumers know if the food they buy is authentic.

Masoud Rostami, associate professor of instruction in data science and assistant director of the Division of Data Science, is principal investigator for the project. The study, “Multimodal Data Fusion and Explainable AI for North Texas Honey Authenticity”, is being funded by a $288,850 grant from the U.S. Department of Agriculture’s National Institute of Food and Agriculture (NIFA).

He is also a co-investigator in collaboration with the UTA School of Social Work for a project to study the reliability, explainability, fairness and potential limitations of machine-learning models trained on federally prosecuted human-trafficking cases.

NIFA food authenticity study

In the NIFA-funded study, Rostami and his team will explore how AI can help to address the increasingly difficult challenge of food authenticity. Food fraud is estimated to cost the global food industry tens of billions of dollars each year.

“Food fraud is a multibillion-dollar global challenge that undermines food supply chain integrity, misleads consumers and threatens consumer confidence, and causes significant economic losses for legitimate agricultural producers,” Rostami said.

Honey is among the foods frequently reported to be adulterated, sometimes through the addition of less expensive sweeteners such as rice, corn and other sugar syrups.

“Every jar of honey carries a natural fingerprint shaped by the flowers bees visit and the environment around their hives,” Rostami said. “Using honey produced in North Texas as a starting point, this research combines biological and chemical evidence with explainable AI to investigate natural signatures, identify potential adulteration, and develop more transparent and reliable authentication methods. The goal is not only to make accurate predictions, but also to explain the evidence behind them.”

The project will combine pollen microscopy and imaging, routine honey chemistry, infrared spectroscopy, and selected advanced chemical analyses with explainable AI to develop more transparent and reliable approaches to regional honey-authenticity screening. The team will create a North Texas Honey Reference Library and a conservative “Pass-or-Review” system that compares samples with verified regional honey, flags those needing closer examination, and explains the evidence behind each flag rather than automatically declaring a sample authentic or fraudulent.

The study has implications far beyond honey, Rostami said. It aims to create a scientific foundation for future AI-powered approaches to detecting food fraud across other agricultural products. The study’s goals also include protecting food producers, strengthening food supply chains, and contributing to more resilient and trustworthy food systems, he added.

Machine learning models study

Rostami’s other new project, titled “Critically Evaluating Machine Learning Models Trained on Federally Prosecuted Human Trafficking Cases”, is funded through a $20,000 grant from the UTA Interdisciplinary Research Program (IRP). It seeks to evaluate the reliability, explainability and potential biases of AI and machine-learning models related to human trafficking cases.

Human trafficking is a global crime involving both sex and labor exploitation. Cases brought to court represent only a subset of trafficking, making prosecution records an incomplete picture of the broader problem.

Rostami and researchers from the UTA School of Social Work will put particular focus on whether models may reproduce institutional patterns and biases present in prosecution-based datasets. The reliability of AI and machine learning predictions depends greatly on the data used to train these models.

“Because prosecution records represent only cases that have been identified and brought to court, AI models trained on these records may reflect patterns in law enforcement and prosecution rather than the broader reality of human trafficking,” Rostami said.

Rostami and social work researchers will combine expertise in data science, social work, and responsible AI and will examine model accuracy, fairness, reliability, interpretability and sensitivity to limitations in the underlying data.

“The goal is to develop a more transparent and scientifically responsible approach to using AI in human trafficking research, helping researchers better understand both the opportunities and limitations of these technologies,” he said. “Trustworthy AI means understanding why a model reaches a conclusion, how uncertain it may be, and when human judgment matters most.”

Taken together, the projects reflect Rostami’s broader research focus on AI that is not only accurate, but also trustworthy, explainable and reliable — particularly when its predictions influence real-world decisions.

---

The UTA College of Science, a Carnegie R1 research institution, is preparing the next generation of leaders in science through innovative education and hands-on research and offers programs in Biology, Chemistry & Biochemistry, Data Science, Earth & Environmental Sciences, Health Professions, Mathematics, Physics and Psychology. To support educational and research efforts visit the giving page, or if you're a prospective student interested in beginning your #MaverickScience journey visit our future students page.