M.S. Data Science (Candidate) | Python & R Developer

This profile serves as the source code repository for my applied research in behavioral analytics and algorithmic forecasting. Unlike my business portfolio, this archive focuses on the raw implementation logic, reproducibility, and statistical methodologies used in my modeling work.

Stack: Python, Scikit-Learn, XGBoost, SHAP

• Implementation: Developed a classification pipeline using scale_pos_weight to handle class imbalance in HR data.
• Logic: Replaced standard black-box predictions with SHAP (SHapley Additive exPlanations) waterfall plots to quantify individual turnover risk factors.
• Key Function: Optimized precision-recall thresholds to improve minority class detection (attrition) from 34% to 61%.

Stack: Python, Facebook Prophet, Pandas

• Implementation: Time-series decomposition model separating growth trends from weekly/yearly seasonality.
• Logic: Calculated dynamic "Safety Stock" levels using 95% confidence interval upper bounds rather than static averages.
• Outcome: Generated risk-adjusted procurement algorithms to minimize stockout probabilities during high-variance periods.

Stack: R, Magick, Tmap

• Implementation: Custom "Manual Device" rendering pipeline to bypass standard library limitations on ARM64 architectures.
• Logic: Script generates individual PNG frames for urban density evolution (1950–2030) and stitches them using the Magick image processing engine.
• Feature: Hardware-agnostic rendering loop for high-resolution time-lapse generation.

Stack: Python, SciPy, Statistical Hypothesis Testing

• Implementation: rigorous evaluation of gameplay gate mechanics (Level 30 vs Level 40).
• Logic: Applied statistical significance testing (p-value analysis) to measure impacts on 1-Day vs 7-Day retention rates.
• Result: Quantified the negative impact of delayed friction points on long-term user habituation.

• Full Case Studies & Business Impact: View Portfolio
• Professional Network: LinkedIn Profile