Developed a machine learning framework to predict socioeconomic development across Connecticut’s census tracts, comparing statistical methods like OLS, random forest, and gradient boosting. The project involved preprocessing socioeconomic indicators, training models, and evaluating their performance to identify the most accurate approach. By analyzing these predictive techniques, the research supports data-driven decision-making in urban planning, investment, and resource allocation for Connecticut’s towns and regions.

I transformed raw socioeconomic data from the U.S. Census into an elegant composite metric using PCA. By extracting maximum variance through eigenvalue decomposition, we bypass arbitrary weighting and derive a statistically robust SES metric. Validation such as correlation and variance analysis helps strengthens the reliability of our approach and aligns with best practices in quantitative social science.

Built a dynamic global trade dashboard in Power BI using DAX expressions and Power Query for advanced data modeling and visualization. Integrated multiple data sources to create interactive reports highlighting trade patterns and key metrics across regions. Populated SQL Server database with 2.1M-row master table across 100 tables with dynamic SQL queries.

This project implements a robust machine learning framework for credit default prediction, leveraging the UCL German Credit Dataset and benchmarking performance against the methodologies proposed in I-Cheng Yeh & Che-hui Lien’s 2009 paper ("The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients"). We extend their work by incorporating modern XGBoost optimization and conducting a rigorous comparative analysis of model performance.