Class Advanced Data Science

'Advanced Data Science' develops a rigorous, end-to-end analytical practice using (mainly) the Ames Housing Dataset as a sustained case study. The module takes students from data characterisation - variable typing, missing-data diagnosis, and quality assessment - through association analysis and collinearity detection, to the construction and evaluation of predictive models. A central portion covers dimensionality reduction (SVD, PCA, and factor models), emphasising correct application within modelling pipelines and interpretation as latent-variable hypotheses. The final weeks address decision-oriented model evaluation, interpretability, and algorithmic fairness. Reproducibility is a cross-cutting concern: students work in Python notebooks throughout, culminating in an integrated reproducible report subject to peer review.

Not applicable

Advanced predictive modelling: regression and classification models with complex datasets and error analysis. Treatment and interpretation of missing data: patterns of absence (MCAR, MAR, MNAR), application of conditional and justified imputation strategies. Data typing and structure: analytical definition of variable types, creation of data dictionaries, and understanding the relationship between encoding and statistical meaning. Feature engineering and dimensionality reduction: application of SVD, PCA, and factor analysis to generate new structured representations. Performance metrics and model explainability (XAI): evaluation with robust metrics (MAE, RMSE, PR-AUC), permutation importance, and SHAP. Algorithmic fairness and bias assessment: fairness (demographic parity, equalized odds), analysis and communication of biases in predictive models. Reproducibility and professional best practices: reproducible projects and reports.  

Interactive peer review: collaborative feedback among students to enhance mutual learning. Immersion sessions: practical sessions that complement theoretical classes. Project-based learning: development of projects covering the entire data science lifecycle. Python notebooks: primary tool for interactive coding and reproducible documentation. Vibecoding (AI-assisted coding from natural-language prompts): applying generative AI tools to enrich practical components and enable rapid iteration on analytical solutions.

Gelman, A., Hill, J., & Vehtari, A. (2020). Regression and Other Stories. Cambridge University Press. Little, R. J. A., & Rubin, D. B. (2002). Statistical Analysis with Missing Data (2nd ed.). Wiley. Note. Students are not expected to buy any books for this module.