• Optimized a database of 10,000 apartment samples, scaled down 100 metrics to 20 to avoid multicollinearity and overfitting• Created XGBoost model segmentation based on key features, successfully eliminated feature importance bias • Implemented cross-validation and hyper-parameter tuning, reduced overfitting by 20% and boosted 10% model performance• Employed SHAP values using beeswarm plot for comparative feature importance analysis across segmentation models• Designed a function that allows users to view individual property, showing a SHAP value waterfall plot and the corresponding real-world value for the top 10 features, enhancing information granularity at the sample level

• Led a team of 4 to analyze enterprises’ fundamentals, financial statements, and SWOT matrix to craft an investment portfolio• Employed Markowitz model, capital market line, and indifference curve methodologies to determine optimal weights of stocks and risk-free assets, tailored to the investor’s risk aversion coefficient • Applied the ARMA model, analyzing historical stock data to accurately forecast future stock price movements, which provided a solid foundation for informed investment decisions and contributed to an annualized portfolio return of 18%

• Analyzed commercial banks’ financial data, including balance sheets, profit statements, cash flow statements, and annual reports• Built a liquidity stress testing framework with three scenarios (mild, moderate, severe) for economic crises simulation in python• Utilized the framework to monitor a series of indexes, including reserve requirement changes and liquidity buffer duration to identify banks’ weaknesses, leading to a 20% improvement in risk mitigation strategies across 4 banks over 2 months