Babak Safaei

Principal Data Scientist (PhD Data Science Intern), Bureau Data Strategy 

Capital One Headquarters, McLean, Virginia | Jun. 2025 – Aug. 2025

Role summary. Served as a PhD data science intern operating in the capacity of a principal data scientist on the Bureau Data Strategy team. Led research and development of an auditable, production-aware feature selection framework for financial technology risk modeling using credit-bureau data and contributed to governance artifacts that calibrate trust in model-assisted decisions.

• Researched feature-importance and explainability methods, including game-theoretic approaches (SHAP, SAGE), gradient-boosted feature gain, and permutation importance, to compare predictive contribution, stability, and runtime efficiency on large-scale credit-bureau data (100M+ records) for financial risk modeling.
• Developed RFE-CV-SHAP, a feature selection framework through recursive feature elimination procedure guided by cross-validated SHAP attributions, to achieve model parsimony while preserving out-of-sample performance and improving interpretive stability for gradient-boosted risk models.
• Operationalized governance by translating explainability outputs into decision artifacts: reason codes and audit trails that enable machine learning engineers and data scientists to trace, challenge, and override model outputs where appropriate.
• Built and validated gradient-boosted machine-learning pipelines for fraud detection and related risk use cases, using XAI outputs to provide interpretable risk explanations for +700k transactions and to support model-governance review.
• Technical stack. Python; PySpark; pandas, NumPy; scikit-learn; XGBoost, LightGBM; SHAP, SAGE; Databricks; Kubernetes; EMP.