Exam score prediction system
MAUK R&D · applied machine learning
From a student dataset to a working predictor: two models compared honestly, and the one that won shipped.
Given a student's study hours, attendance, sleep, previous grade and background, this predicts their final exam score. Two models were trained and compared; the simpler one won, and predictions run instantly in the browser at no hosting cost. The comparison stays in the interface, so the losing model is visible too.
- What it is
- Web tool
- Industry
- Education
- Status
- Working, not public
- Year
- 2026
The problem
What needed solving
Small data problems tempt teams into heavy models. The question worth answering was whether a more complex model actually earns its place here, and how to serve whichever one wins without paying for a server.
The solution
What we built
A prediction form backed by the better of two trained models, with the evidence for that choice kept on the same page.
Core functionality
- Nine inputs covering study habits, background and previous grade
- Instant predicted score out of 100
- A comparison table of both trained models
- Downloadable charts, comparison data and the trained model
Screens
More screenshots
Captured from the live site, a demo workspace, or the product running on sample data. No customer records, passwords or private admin data are shown.
Business impact
What it achieved
0.617
fit on students the model never saw (R²)
Against 0.576 for the more complex model.
Verified±4.98
average error in exam points
Verified1,000
student records used to train and test
Verified
What sets it apart
Why it works
- 01
The simpler model won
The straightforward model beat the more complex one on every measure, so it shipped. Complexity has to earn its place.
- 02
No server to pay for
Predictions are calculated instantly in the browser, so the tool costs nothing to host.
- 03
The evidence stays visible
Both models' scores and charts sit beside the predictor, so anyone can check the choice.
Services
Where this work fits
Related projects
Similar projects
LiveIn-house buildVerifiedMulti-disease risk screening model
MAUK R&D · applied machine learning
Risk scores for ten conditions from 25 routine measurements, trained on 281,000 records and private enough to run on the patient's own device.
86.88%accuracy on patient records the model never saw
Machine learning
View project
LiveIn-house buildVerifiedSpamShield email classifier
MAUK R&D · applied machine learning
A spam classifier built from first principles that explains every verdict it gives.
83.8%accuracy on 920 emails the model never saw
Machine learning
View project
ResearchVerifiedRoad-scene object detection on BDD100K
MAUK R&D · computer vision
A road-scene detection model trained from scratch on 70,000 images with a laptop GPU, and an honest account of where it struggles.
0.575detection accuracy (mAP50) on 10,000 unseen images
Machine learning
View project
Discuss a similar project
Building something like Exam score predictor?
We'll tell you what we'd reuse, what we'd change and what it would take.