Nigeria and PhD students in applied mathematics at Deborah Okori, Mississippi State University, have built an easy-to-understand machine learning system for the online marketplace.
In an interview with Punch on Tuesday, she said her research focuses on helping businesses and policymakers make smarter, data-driven choices in a fast-paced world of e-commerce.
“My goal is simple for leaders to build machine learning models that they can question and trust. It's a model that not only gives answers, but also explains how they got there,” she said.
Okoli's work focuses on what she calls Lag-Aware machine learning.
She said her research examines economic factors such as labor productivity, R&D, sales, employment and capital investment.
Rather than relying on black box algorithms, she designs time-shifted features and applies a model that explains both the intensity and timing of the effects of each factor.
She said: “I use a variety of models, from normalized regression to tree-based algorithms.
“Next, we apply explanability tools such as feature attribution and partially dependent plots.
“These techniques reveal how each variable affects predictions and whether its impact is immediate or delayed.”
She added that all models had been rigorously tested before sharing the results.
“We use cross-validation of rolling windows to simulate real-world conditions, stability checks for the entire time lag, and residual diagnosis to ensure that there are no missing patterns in the model,” she explained.
Her results were also compared to traditional economic baselines, she said.
“This is important to ensure that machine learning is actually adding values.
According to her, the final outcome is a clear prediction with estimated confidence ranges and simple interpretations.
Okoli said, “For example, a surge in productivity could mean an increase in online sales for the next quarter, but the impact of R&D spending could unfold more slowly. These insights allow businesses to plan their inventory, coordinate their logistics and make aggressive digital strategy decisions.”
She also emphasized that her work was built on openness.
She said, “A prediction should come with a seat belt. If the outcome can change dramatically due to one new data point, decision-makers need to know that in advance.”
Born and raised in Nigeria, Okoli graduated from the top honors of industrial mathematics at Covenant University in Ogun State and she was the best student in her entire department.
She began her graduate studies in applied mathematics at Tennessee Institute of Technology before transferring to Mississippi State University where she is currently earning her PhD.
She also holds a Masters degree in Educational Research from Hull University, UK.
In Mississippi, he works under the leadership of Professor Kim Sung-jai of the Department of Mathematics and Professor Jason Singh of the University of Business, and has developed reproducible, adaptive forecasting templates that allow organizations to tailor their own economic data.
Looking ahead, Okori said she wanted to broaden her research scope.
“We want to create plain language prediction tools for non-technical users, develop AI templates that can be adopted by small and medium-sized businesses and institutions, and promote open collaboration that will increase the standards of clarity and accountability in machine learning.
“As e-commerce continues to grow, it's not enough to predict demand. You need to understand that. Machine learning becomes easier to understand and you can trust it. And when it's reliable, it becomes useful,” Okori said.
