✦ AI Use Case

Student Retention Predictive Analytics
Early Warning System with Azure ML

How an Azure Machine Learning model integrated into Power BI gave academic advisors an early-warning system that identified at-risk students by week 4 — enabling proactive intervention weeks before grades failed.

🏚 Industry: Higher Education
🛠 Tools: Azure ML, Power BI, Copilot, Power Automate, SQL Server, Azure Synapse
Impact: At-risk students identified 6–8 weeks earlier

Business Problem

Scenario: Late Identification of At-Risk Students

Academic advisors only became aware of struggling students after mid-term grades were posted — usually 10–12 weeks into the semester. By that point, many students had already disengaged and withdrawal was often the only remaining option. The university had rich historical data but no proactive system to act on it.

AI-Powered Approach

Step 1 — Rich Student Feature Dataset

Built a weekly feature dataset in Azure Synapse combining attendance, assignment completion, grade trends, financial aid status, and prior GPA to feed the predictive model.

Step 2 — Azure ML Retention Model

Trained a classification model in Azure Machine Learning on 5 years of historical data. The model predicts the probability of student withdrawal or academic failure by week 4 with 81% accuracy.

Step 3 — Real-Time Score Integration

Published the Azure ML model as a real-time endpoint and embedded weekly risk scores directly into the Power BI advisor dashboard, along with key contributing factors.

Step 4 — Advisor Action Dashboard with Copilot

Created a Power BI dashboard that sorts each advisor’s caseload by risk score and uses Copilot to generate personalized student summaries with recommended interventions based on historical success patterns.

Step 5 — Automated Alerts via Power Automate

Configured Power Automate to send immediate Teams notifications to advisors when any student crosses a risk threshold, including a direct link to the student’s profile and suggested outreach talking points.

Key Benefit: Advisors can now intervene proactively in weeks 4–6 instead of reacting in weeks 10–12.

Measured Outcomes

81%
Prediction accuracy by week 4
65%
Increase in early intervention rate
Week 4
At-risk identification (vs. week 10–12)
Improved
First-year retention rate within 2 years