Machine Learning Projects
Neural-network classifier with proper evaluation
Coursework, NUST · Developer · 2025
A small but complete supervised-learning project: train/validation/test splits, early stopping, ROC curves and confusion matrices on every split, and a regression baseline. Included here because it shows how we evaluate models, not just train them.
The problem
Predict student outcomes from a tabular dataset and report results honestly, with a held-out test set that was never used for training.
What was done
- ✓Separate training, validation and test sets with early stopping on validation checks
- ✓ROC curves and confusion matrices reported per split, not just overall accuracy
- ✓Error histogram and a linear-regression baseline for comparison
Evidence
Artefacts
- Training curves
- ROC per split
- Confusion matrices per split
- Regression baseline
Outcome
A model report a client can trust: per-split metrics and failure cases, not a single accuracy number.