MCM/ICM Mathematical Contest in Modeling 2025

Project Overview

International mathematical modeling competition tackling an open, data-driven problem. I assembled multi-source datasets, framed hypotheses, and built a transparent pipeline—from exploratory analysis and feature engineering to model selection and visualization—then translated results into an actionable narrative and memo.

What I Did

• Scoping & data assembly: curated public datasets; standardized country names/years; reconciled missing values.
• EDA & feature engineering: trend and seasonal checks; log/ratio transforms; composite indicators; outlier and leverage diagnostics.
• Modeling: baselines (OLS, regularized regression) → tree ensembles for nonlinearity; sensitivity checks and ablation to avoid over-fitting.
• Clustering & segmentation: region- and readiness-based cohorts to compare dynamics and policy efficacy.
• Forecasting & what-ifs: simple time-series baselines with scenario knobs to stress-test policy levers.
• Visualization: heatmaps, small-multiples, and tier maps for cross-country comparability; reproducible charts for the paper.
• Communication: executive summary + appendix (assumptions, limitations, reproducibility notes) tied to recommendations.

Reflection

Biggest lesson: modeling is only useful when the assumptions, comparability, and uncertainty are explicit. Cross-country data is noisy and policy signals are confounded; versioned data cleaning, clear definitions, and sensitivity analysis matter as much as the final metric. Packaging results as a short policy story—backed by transparent exhibits— made the work land with non-technical readers. If iterating, I’d tighten causal identification (instrumental variables or synthetic controls where feasible) and expand counterfactual scenarios to pressure-test recommendations under data drift.