Data Science Stuff

## about

Data science project collection showcasing different ETL and ML techniques on numerous different datasets.

Data Science Stuff is the collected work from my Data Science Fundamentals course. It includes smaller studies on penguins, life expectancy, breast-cancer data, PCA, k-nearest neighbors, and gradient descent, plus a larger capstone on academic performance.

The capstone used a UCI dataset with roughly 145 students and 32 demographic, socioeconomic, and academic features. The sample is small, so I treated the project as an exercise in careful modeling and interpretation rather than a basis for claims about students in general.

## capstone

I tested DBSCAN and k-means for student grouping. DBSCAN collapsed to one cluster, and the best k-means result used two clusters with a weak silhouette score near 0.11. That was a useful result in itself: a clean-looking PCA plot did not mean the clusters were especially strong.

For grade prediction, I reduced the sparse eight-class target to low, middle, and high groups. A tuned random forest reached about 64% test accuracy. Permutation importance and SHAP pointed toward prior academic performance, study time, age, reading habits, and family context, but I kept the write-up correlational and avoided turning a small observational dataset into a causal story.

## other work

The rest of the collection covers k-nearest-neighbor classification on Palmer Penguins, country clustering and PCA on WHO life-expectancy data, PCA and classification on the Wisconsin breast-cancer dataset, and a manual gradient-descent implementation with interactive Plotly traces. A separate ethics paper looks at privacy, governance, and the risk of repeating weak data claims until they sound true.

## screenshots

no images — this one lives in the code.