Sangram Keshari Patro

I’m Sangram Keshari Patro, a results-oriented data scientist with a strong foundation in computational physics and a passion for solving real-world business problems through data. With an M.Sc. in Physics from UM-DAE Centre for Excellence in Basic Sciences and a PGP in Data Science & Business Analytics from the University of Texas at Austin (expected Aug 2025), I bring both analytical depth and practical experience in predictive modeling, machine learning, and statistical analysis.

Throughout my journey, I’ve worked on impactful end-to-end projects across industries:

• Churn Prediction for a DTH Provider: Built a machine learning model to flag high-risk customer accounts and recommended targeted campaigns, balancing cost and retention potential.
• INN Hotels Booking Cancellation Prediction: Developed a logistic regression and pruned decision tree model (F1 score: 0.80) using 36K+ records. Insights on lead time and special requests led to strategies with 22% projected reduction in cancellations.
• OTT Viewership Forecasting – ShowTime: Created a linear regression model (Adj. R²: 77.4%) identifying that 6.5M more platform visitors drove 1M extra views, while sports events reduced views by 63K.
• ABC Estate Wines Sales Forecasting: Applied ARIMA and triple exponential smoothing (RMSE: $9,334) on 30+ years of data to capture Q4 demand peaks and guide festive-season inventory and marketing.
• Visa Approval Optimization – EasyVisa: Tuned an XGBoost classifier (F1: 0.825) with SHAP explainability, finding every $10K salary bump increased approval odds by 1.3%.
• Customer Segmentation – AllLife Bank: Applied K-Means and PCA to 600+ profiles, segmenting users into actionable clusters with up to 2.5x credit limits.
• Sales Drop Analysis – New Wheels: Used advanced SQL on messy transaction data to detect a 35.8% sales drop, 47% spike in negative ratings, and 1.16-point satisfaction dip, guiding pricing and feedback loops.

During research internships at TIFR, I modeled lifetime extractions using chi-square fits and performed detector efficiency comparisons using Python, C++, and Radware

🛠 Skills: Python, SQL, C++, MATLAB, Tableau, Power BI, KNIME, Shell scripting, LaTeX

📊 Tools: Scikit-learn, Statsmodels, Seaborn, Shap, Pandas, Numpy, Plotly, Radware

🏆 Achievements: Top 3 - Great Learning Hackathon, Top 4/1000 - Analytics Vidhya AI Battlefield, DAE DISHA Scholar (Top 50 of 40K), NSTSE AIR 104

I'm always eager to work on data-driven initiatives that create measurable impact across domains.