Therapy Dropout Prediction (Humana Mays Healthcare Analytics Competition)

• Conducted EDA on 500k+ Rx/medical claims data with best practices to find predictive features leading to premature therapy end.
• Engineered features in Python to achieve a 0.94 AUC score with XGBoost, implemented logistic regression to reveal primary dropout predictors, and predicted the treatment discontinuation risk for every 1232 patients following an agonistic drug event.

Service Access Web App (UPS Technology Group Hackathon, Top 5/50)

• Designed a Streamlit-based web app with assessment, query features, and employee authentication by building a CatBoost model
• Predicted employee access to resources, reduced manager check time by approx. 20%, and raised prediction accuracy by 5% by hyper-tuning the learning rate, iterations, and depth of gradient-boosted trees.
• Provided presentation to management on how the app supports reducing onboarding wait times and offboarding audits.

Financial Insight Web App

• Built a Tableau dashboard by integrating portfolio optimizer and sentiment analysis with risk assessment for informed decision- making, as well as visualized historical company performance with price and volume charts.
• Performed web scraping of 4 years of financial news & reports using Beautiful Soup, FINBERT, and tokenization in NLP. ▪ Analyzed & recommended 18 stocks with high returns & minimal risks with Monte Carlo simulation and Sharpe Ratio.
• Analyzed & recommended 18 stocks with high returns & minimal risks with Monte Carlo simulation and Sharpe Ratio.

Best Buy Daily Sales Forecasting (MSA Project sponsored by Best Buy)

• Used Pandas, NumPy, and Matplotlib for data preprocessing, data transformation, and data visualization of 5-year sales data.
• Performed feature preprocessing and selection using PCA, Random Forests, correlation matrix, and one-hot encoding.
• Implemented XGBoost and Facebook Prophet models to enhance sales forecasting & performance and reduced RMSE by 50%.

Movie Recommendation System

• Explored ~800k movie rating data and conducted descriptive data analysis using Python Pandas & Seaborn visuals.
• Transformed genre into 32 dummy features, identified similarities between movies based on user & movie features using hierarchical clustering with unevenly distributed genre data, selected silhouette coefficient to optimize cluster coherences.
• Implemented MLP with SGD & ADAM optimization, SVD, SVD++, KNN, Bernoulli & Gaussian Naive Bayes classifiers, fined-tuned model hyper-parameters and achieved 67.7% accuracy to predict if a user would like a movie using SVD++.

Data Science Club - BirdCLEF Team

• Led spatial autocorrelation analysis with ArcGIS using data science and manipulation techniques to find the populated regions.
• Developed lightGBM for predicting bird species occurrences based on spatial and temporal data to reveal migration patterns.
• Automated statistical & spatial workflows, enhanced data processing, and visualized geospatial data with Geopandas and PyTorch.

Cancer Survival Prediction - 2019 Hackathon Hacklytics

• Led a team of 3 analysts to build a linear regression model from Scikit-learn in Python to analyze the correlation between number of mutations per gene and estimated survival rate among cancer patients, with an adjusted R-squared value of 0.76.
• Showcased analytical findings, trends & correlations between mutations & percentage of cases affected, derived personal mutation risk based on 5 demographic parameters using an interactive Tableau dashboard.