Student Success Predictor

Predictive analytics system to identify at-risk students and improve retention rates.

Published March 1, 2024

1 min read

Technologies Used

Python

Canvas LMS API

Tableau

PostgreSQL

Scikit-learn

Dashboard showing student risk levels and intervention recommendations

Saving Students Before They Slip Away

Dropout rates in higher education are a persistent challenge. Traditional early warning systems often rely on mid-term grades, which is too late for effective intervention. We built a Student Success Predictor that analyzes real-time behavioral data from the Learning Management System (LMS)—login frequency, forum participation, assignment submission times—to flag at-risk students weeks before their grades start to suffer.

The Challenge

Reactive intervention strategies based on lagging indicators (grades).

Siloed data between the Registrar, LMS, and Student Services.

Generic support emails ignored by struggling students.

Limited resources for academic advisors to track hundreds of students.

The Solution

Ingested real-time activity streams from Canvas LMS via API.

Developed a Random Forest model to predict dropout probability.

Created a prioritized 'Intervention List' dashboard for advisors.

Automated personalized check-in emails based on risk triggers.

Key Features

Risk Radar

Visual heatmap of student engagement across all courses.

Login velocity
Submission latency
Peer interaction score

Advisor Dashboard

Centralized view for advisors to manage caseloads and track interventions.

Risk sorting
Note taking
Meeting scheduler

Impact Analysis

Measure the effectiveness of different intervention strategies.

Retention lift
GPA improvement
Resource utilization

Technology Stack

Python

Canvas LMS API

Tableau

PostgreSQL

Scikit-learn

Project Timeline & Results

Week 1-3

Data Pipeline

Connected to Canvas LMS and SIS (Student Information System).

Set up secure ETL pipelines to extract daily activity logs. Merged LMS data with demographic and historical academic records from the SIS. Anonymized PII for model training.

Week 4-6

Model Development

Achieved 78% recall in identifying dropouts.

Engineered features like 'days since last login' and 'forum post sentiment'. Trained the predictive model on historical dropout data. Validated against the previous semester's outcomes.

Week 7-9

Pilot Program

Advisors reported 2x efficiency in identifying students in need.

Rolled out the dashboard to a pilot group of academic advisors. Refined the risk thresholds based on their feedback. Automated the weekly email digests.