A New Approach to Machine Learning in E-Commerce
Deborah Okoli, a Nigerian PhD student in Applied Mathematics at Mississippi State University, has developed an innovative machine learning system designed to make data-driven decisions more accessible for businesses and policymakers in the fast-paced world of e-commerce. Her work focuses on creating models that not only provide accurate predictions but also explain their reasoning, making them more trustworthy and actionable.
Okoli’s research centers around what she refers to as “lag-aware machine learning.” This approach goes beyond traditional methods by considering not just what drives online market growth, but when those factors have an impact. By incorporating time-shifted features into her models, she is able to capture the delayed effects of economic variables such as labor productivity, research and development (R&D), sales, employment, and capital investment.
Instead of relying solely on complex black-box algorithms, Okoli uses a variety of models, including regularized regressions and tree-based algorithms. She then applies explainability tools like feature attribution and partial dependence plots to reveal how each variable influences the forecast and whether the impact is immediate or delayed. This level of transparency allows decision-makers to understand the reasoning behind the predictions, which is crucial for making informed choices.
Before sharing any results, Okoli ensures that every model undergoes rigorous testing. She employs techniques such as rolling-window cross-validation to simulate real-world conditions, stability checks across time lags, and residual diagnostics to identify any missing patterns. Additionally, she compares her findings with traditional economic baselines to confirm that the machine learning models are adding value rather than simply generating numbers.
The final outcome of her work is a clear forecast accompanied by an estimated confidence range and a straightforward interpretation. For instance, a spike in productivity might indicate increased online sales in the next quarter, while the effects of R&D spending may unfold more gradually. These insights enable businesses to plan inventory, adjust logistics, and make proactive digital strategy decisions.
Okoli emphasizes the importance of openness in her work. She believes that forecasts should come with “seatbelts” — if a new data point could drastically change the outcome, decision-makers need to be aware of this upfront. This philosophy underpins her commitment to building models that are not only accurate but also transparent and interpretable.
Born and raised in Nigeria, Okoli graduated with First-Class Honours in Industrial Mathematics from Covenant University in Ogun State, where she was the top-performing student in her department. She began her graduate studies in Applied Mathematics at Tennessee Tech University before transferring to Mississippi State University, where she is currently pursuing her PhD. In addition to her doctoral studies, she holds a Master’s in Education Research from the University of Hull in the United Kingdom.
At Mississippi State, Okoli works under the mentorship of Professor Kim Seongjai from the Department of Mathematical Sciences and Professor Jason Shin from the College of Business. Together, they are developing reproducible and adaptable forecasting templates that organizations can customize using their own economic data.
Looking ahead, Okoli aims to expand the reach of her research. She hopes to create plain-language forecasting tools for non-technical users, develop AI templates that small businesses and agencies can adopt, and promote open collaborations that enhance clarity and accountability in machine learning. As e-commerce continues to grow, she believes it is no longer enough to predict demand — understanding it is equally important. When machine learning becomes easy to understand, it becomes trustworthy, and when it is trustworthy, it becomes useful.
