Project Notes
Six classifiers, one medical dataset
What comparing several classification approaches taught me about model evaluation in a medical-data setting.
Running several algorithms on the same dataset is easy. Learning something useful from the comparison requires more discipline.
The project
I compared six classification approaches on breast-cancer data using cellular characteristics to distinguish malignant and benign cases. The exercise became less about finding a favourite algorithm and more about understanding how preprocessing, model assumptions and evaluation choices shape the result.
The main lesson
In medical-data work, a single accuracy number is rarely the whole story. Error types matter, validation design matters, and the apparent simplicity of a tabular dataset can hide important modelling assumptions.
Why I keep this project in the portfolio
It is an early project, but it shows a useful progression: moving from implementing algorithms to asking better questions about comparison and evaluation. That progression matters more to me than presenting one model as universally superior.