Student Performance Prediction System source code

The problem it solves

Lecturers need a reliable way to identify students at risk of failing or underperforming early, using stored marks and attendance, while students need private feedback on their predicted outcome. Manual review of records is slow and inconsistent.

Project topic: Student Academic Performance Prediction System Using Machine Learning

Who uses it

What it does

• Login and role access

  Standard login with hashed passwords and role-based redirects.

• Student record management

  Form to add one record and CSV upload with line-numbered error reporting.

• Prediction list and student detail

  Filtered list sorted with at-risk students first, with drill-down to confidence and top factors.

• Model training and metrics

  Trains decision tree on held-out split, shows accuracy, precision and recall, and retrains on demand.

• CSV export and printable report

  Download filtered predictions as CSV and view a printer-friendly class report.

• User and course administration

  Manage accounts and course lists.

Screenshots

Screenshots of the running system, taken by our checks. The records are sample data and are labelled so on each page.

Database tables

• usersLogin accounts and role assignments.id, username, password_hash, full_name, role, created_at
• studentsStudent personal details.id, matric_no, name, level, user_id
• coursesCourse catalogue.id, code, title, unit, lecturer_id
• enrolmentsOne row per student per course per semester.id, student_id, course_id, semester, attendance, assignment_avg, test_score, exam_score, final_grade_band
• predictionsPrecomputed predictions and explanations for display.id, enrolment_id, predicted_band, confidence, top_factors, created_at
• model_metricsLatest evaluation metrics for display.id, accuracy, precision, recall, trained_at

Technology

• Python. Required by the kit and suitable for the machine learning code.
• Flask. Lightweight and easy to explain, with one blueprint per feature area.
• SQLite through sqlite3. No server to set up and it works fully offline.
• Plain Python decision tree in app/model.py. Keeps the project self-contained and offline, and the tree rules can be read out at defence.
• Jinja templates and Bootstrap locally. Responsive pages that work on a projector and a phone, with all files held in app/static/vendor.
• Werkzeug security and Flask-WTF CSRF. Kit defaults for security, used on every form.

How to run it: running a Python project on Windows. Not sure which language to pick? Python or PHP.

How it was tested

31 of 31 tests passed, 24 of them automated. The system was also installed from scratch and every screen was opened as the right kind of user before it was accepted.

What it does not do

• The sample data is small and fictional, so the reported accuracy, precision and recall describe the sample only and are not a claim about real students.
• The decision tree uses fixed hyperparameters, a fixed maximum depth of four and a fixed random split, so results are repeatable but not tuned.
• Predictions are estimates and are never official grades. Every screen that shows a prediction carries a disclaimer.
• The plan says the administrator retrains on demand, so a prediction is not refreshed automatically after a single new record.
• A row cannot be edited or deleted after upload, so a mistake is corrected by reseeding or by adding the correct semester instead.
• The model trains on the final grade band already stored in the enrolments table, so the labels come from the same marks used as inputs.
• The interface is a responsive web page only. There is no mobile app.