End-to-End ML Practice
A working set of smaller exercises that deliberately trade depth for breadth — each one runs the full lifecycle (EDA → modeling → evaluation) on a different problem type.
The problem
Alongside the flagship case studies, this repository holds a broader practice set: bike-sharing demand regression, HR attrition classification, credit-risk scoring, diabetes-risk prediction, restaurant and product-review sentiment analysis, and news-article categorization, among others.
Approach
Each notebook follows the same disciplined shape — exploratory data analysis, missing-value and outlier handling, feature engineering, model comparison across multiple algorithm families (linear models, tree ensembles, clustering), and a final evaluation against the metric that actually matters for that problem (RMSE for regression, F1/recall for imbalanced classification, silhouette score for clustering).
Representative techniques
- Regression: Ridge/Lasso regularization, bike-sharing demand forecasting
- Classification: HR attrition, diabetes risk, German credit scoring
- NLP: product-review and restaurant-review sentiment via word embeddings, news-article categorization
- Unsupervised: clustering and segmentation exercises
Breadth practice like this is what makes the flagship projects faster to build well — the same EDA-to-evaluation discipline shows up in EasyVisa and ReneWind because it was rehearsed here first, on a dozen smaller problems.