Parker Hayashi

Duke Impact Investing Group · Durham, NC

• Aggregate user-level behavioral and NLP-derived features from checkout activity using pandas and NumPy
• Engineer churn-related metrics (e.g., cancellation ratios, engagement rates, sentiment aggregates, and LLM-based risk scores) and visualize distributions using matplotlib
• Integrate behavioral signals with aggregated textual features and develop modular Python scripts for automated EDA and feature aggregation to support downstream modeling

Machine learning pipeline that predicts professional ATP tennis match outcomes from decades of historical data.

• Developed a machine learning model to predict ATP tennis match outcomes using 50+ years of match data
• Engineered 100+ features including Elo ratings, surface stats, and head-to-head history
• Achieved high prediction accuracy via XGBoost with cross-validation and isotonic calibration

Source

Full-stack dashboard for macroeconomic indicators with data from the Federal Reserve Economic Data (FRED) API.

• Developed a full-stack economic dashboard with React for interactive visualization of 500+ economic time series
• Engineered a scalable FastAPI backend to retrieve and serve data from the FRED API
• Implemented client-side filtering, multi-series comparisons, and adaptive layouts using React hooks, conditional rendering, and Flexbox/Grid

SourcePreview

Multimodal machine learning framework for overall survival prediction in non-small cell lung cancer (NSCLC).

• Developed a multimodal XGBoost-based model to predict NSCLC overall survival using 36 NLP-processed clinical and genomic features
• Achieved up to 0.84 AUROC, outperforming previous state-of-the-art NSCLC survival models by up to 16.67%
• Provided a tool to help physicians personalize treatment strategies and improve patient quality of life
• Awarded 1st Place Overall and Human Health Category at the 2024 International Forum on Research Excellence

Source