The 30-second version
I built a food-item classifier and compared decision trees, K-nearest neighbours, and neural networks under the same prediction task. The work gave me an early way to reason about feature choices, overfitting, and generalization across model families.
What I can support from the current record
- Python implementation and feature engineering for a structured classification task.
- A comparison of decision-tree, KNN, and neural-network approaches.
- Analysis of training behavior versus held-out behavior, rather than choosing a model from training performance alone.
What is intentionally not claimed
The original dataset description, split procedure, baseline, per-class results, and final metric table are not attached to this portfolio. Without those artifacts, I cannot make a responsible performance claim or call one model “best.” This page therefore remains a short course-project record rather than a polished machine-learning case study.
What I would require in a second pass
Before expanding this page, I would recover the report or notebook and document the data source, sample and class balance, leakage checks, train/validation/test procedure, baseline, and per-class error pattern. Those details—not a longer technology list—would determine whether the result is worth interpreting.