Business Intelligence  |  Predictive Modeling |  Higher Ed Data Strategy

Predicting the one-year retention of students is crucial for any university's intervention strategies concerning at-risk students. At Boise State University, existing models for this purpose were outdated. I managed to enhance the predictive capability of these models by 27% through engineering several new features representing indirect measures of student motivation. (These features could be derived from existing data within the university's source systems.) Additionally, I developed processes to gather supplementary information from students through a new survey to further enhance the accuracy of the models.

Some of this work was used to help inform the development of Oracle's student retention functionality in their student success platform as part of a collaboration between the university and Oracle product developers.

I led a 6-month implementation of four XGBoost predictive models to identify students at risk of not returning the following semester or graduating on time, impacting approximately 20,000 students each term. This enterprise-level solution expanded upon previous work, achieving strong predictive performance with AUC scores ranging from 0.85 to 0.91. Built using Salesforce Einstein Analytics, the implementation includes interactive dashboards and integrated scorecards that deliver actionable insights, enabling more targeted and effective student intervention strategies across the institution.

Boise State University's advising unit for online program students faced a challenge: essential student information was scattered across various source systems and individual Excel files. I did the data warehousing, semantic model creation, and dashboard development work needed to get all the data advisors needed in one place. This project involved closely examining advisors' real-time interactions with student data to ensure the dashboard's design aligned with their processes, facilitating accessibility and user-friendliness.

The Power BI work above was undertaken to support an application designed to monitor student peer assessments within nursing and MBA programs. Beginning with disorganized flat files, I imported them into SSMS for thorough cleaning before constructing a star schema data model to facilitate reporting. An essential project objective was to utilize the data to pinpoint students who were not adequately participating in peer discussions, assessed against predetermined criteria and compared to their peers' performance.

This dashboard was created to facilitate a detailed graduating student survey that I developed. The survey was intricate, featuring extensive branching logic based on embedded demographic data and respondent choices. The most demanding aspect of the project was meeting the diverse expectations of college deans and VPs. Additionally, a significant challenge involved educating users about prioritizing the survey's margin of error and potential response bias over the response rate when evaluating data quality and how to make informed decisions given the often complex patterns in the data.