Foundations of statistics and classical learning

What they are

Four assignments from the statistics part of my double degree, each one about a different family of problem:

Supervised classification. Comparing classification algorithms and justifying the choice by their behaviour, not by their popularity.

Unsupervised segmentation. Clustering a real voting-intention survey published in November 2023. The interesting part is not the algorithm, it is that every cluster has to be explainable: a partition that is coherent according to the metric and makes no political sense is not acceptable.

Regression. Predicting the price of a home and analysing which variables carry the explanatory power.

Multivariate analysis in R. Studying normality, outliers and latent structure on a housing database, applying principal components and factor analysis.

Why I include them

Because they are the foundations, and a data science portfolio that only shows neural networks and voice agents is telling half the story.

Everything I do afterwards rests on this: knowing whether a variable is normal, why correlation does not imply structure, what it means for a component to explain 60 % of the variance, and why one outlier can ruin a model trained on squared error.

What they are not

They are not projects with users. They are academic deliverables with their problem statement, their deadline and their report, and I say so on the card itself because a recruiter who opens them is going to see a PDF and a list of columns.

One of them is not even code: they are KNIME visual workflows. It is a legitimate tool in analytics and a deliverable that cannot be reviewed by reading, which is exactly why I prefer notebooks when I have a choice.