Open University Student Retention & Performance Analysis

Executive Summary

This project analyzes the Open University Learning Analytics Dataset (OULAD) using PostgreSQL to uncover the underlying drivers of student retention, demographic performance gaps, and the impact of virtual engagement. By engineering a relational database and querying over 10 million rows of behavioral logs alongside academic records, this analysis provides actionable, data-driven insights to help educational stakeholders identify at-risk students and standardize curriculum difficulty.

Tech Stack & Skills Highlighted

• Relational Database: PostgreSQL (v18.3), pgAdmin 4
• Query Language: SQL
• Techniques: Common Table Expressions (CTEs), Window Functions (NTILE, PARTITION BY), Conditional Aggregations (CASE WHEN), Multi-table JOINs, Data Type Casting.

---

Core Business Insights

1. The Behavioral Cliff (Digital Engagement as a Leading Indicator)

There is a massive, quantifiable drop-off in digital engagement between passing and failing cohorts.

• Students who Pass interact with the virtual portal an average of 1,922 times.
• Students who Fail average only 688 interactions.
• Business Value: Students who eventually withdraw average a dismal floor of 445 clicks. This establishes digital interaction rates as a highly viable leading indicator, allowing the university to trigger automated intervention emails for low-engagement students weeks before they actually fail an exam.

2. Grading Discrepancies & "Weed-Out" Courses

Using Window Functions to isolate the Top 10% of students per module revealed severe inconsistencies in grading curves across the university.

• Module AAA is graded on a strict, balanced curve: a score of 81.80 places a student in the elite Top 10%.
• Module CCC is highly polarizing. It has the highest withdrawal rate in the university (~46%), yet the surviving cohort scores exceptionally high, requiring a massive 92.50 just to break into the Top 10%. This indicates a severe skill-bifurcation in the curriculum that needs departmental review.

3. Demographic Performance Gaps

There is a strict, linear correlation between a student's prior education level and their final academic performance. Average assessment scores step down sequentially from Post Graduate (83.49) to No Formal Qualifications (70.60). This establishes highest_education as a statistically significant feature for any future predictive modeling regarding student success.

---

Data Architecture & Engineering Challenges

The database was locally architected using 7 interconnected tables. Working with raw, uncleaned Kaggle data presented several real-world data ingestion challenges:

• Handling Nulls & Strict Constraints: The raw CSV data contained thousands of empty text strings ("") in numeric columns (such as dates and scores). To prevent PostgreSQL COPY import failures, tables were dynamically rebuilt using VARCHAR(20) to successfully ingest the dirty data. UPDATE and ALTER TABLE scripts were then executed to cast empty strings to true SQL NULL values, restoring mathematical INT/FLOAT integrity for analysis.
• Primary Key Workarounds: During the import of demographic data, strict PRIMARY KEY constraints flagged false duplicates (e.g., a single student enrolling in multiple semesters). Constraints were temporarily dropped to allow full ingestion, relying on composite grouping (id_student, code_module, code_presentation) downstream to maintain the correct table grain.
• Referential Integrity Validation: Executed LEFT JOIN anti-join patterns between the transactional assessment table and demographic dimension table, verifying 0 orphaned records prior to executing the core analysis.

---

Repository Structure

• /Queries/01_database_setup_and_cleaning.sql: The DDL/ETL script detailing table creation and data type casting.
• /Queries/02_exploratory_data_analysis.sql: Queries validating table grain, missing data volume, and categorical baselines.
• /Queries/03_core_business_analysis.sql: The primary analytical script containing CTEs and Window Functions.
• /Outputs/: Folder containing the exported CSV data outputs for the four core business questions.