Selected portfolio

Four projects. One professional direction.

Each case study connects a clearly defined business problem to data preparation, predictive modeling, validation, and strategic recommendations.

PROJECT 01 · FINANCE + NLP

Stock Return Prediction Using SEC Filing Text

Compared financial-only, text-only, and combined feature pipelines to test whether filing language adds out-of-sample return signal.

OLSNLPTime-aware validation

PROJECT 02 · HEALTHCARE

Early Diabetes Risk Prediction

Developed classification models using CDC BRFSS health indicators to support earlier preventive intervention and targeted outreach.

Logistic RegressionRandom ForestXGBoost

PROJECT 03 · HEALTHCARE FINANCE

Predicting Preventable High-Cost Patients

Used MEPS longitudinal data to identify patients likely to become high-cost users before crisis care.

ROC-AUC 0.8174.31× liftRisk stratification

PROJECT 04 · FORECASTING + OPTIMIZATION

Optimizing Hospital Resource Allocation

Combined next-week occupancy forecasting with linear optimization to recommend staffing and surge-bed decisions.

R² 0.907RMSE 2.888Linear optimization

Interested in a private project briefing?

Complete technical materials are not publicly distributed. Recruiters, collaborators, and organizations may request a confidential discussion about methodology, findings, and implementation.

Request a Confidential Briefing

How the work was developed

Each project follows a complete analytics lifecycle.

The case studies are intentionally more detailed so visitors can understand the depth of the work without accessing protected project files.

PROJECT 01 · STOCK RETURN PREDICTION

Financial text, quantitative metrics, and time-aware validation

This project tested whether sentiment, tone, readability, and other numerical text features extracted from SEC filings add predictive value beyond traditional financial metrics.

Separated financial-only, text-only, and combined feature sets
Applied chronological validation and a three-month buffer
Used OLS as an interpretable benchmark
Compared prediction error, direction, and rank quality

Business questionDoes filing language improve return prediction?

Data challengeAvoiding leakage in forward-looking returns

Primary valueTransparent evidence rather than exaggerated claims

Portfolio lessonInterpretability and honest validation matter

PROJECT 02 · DIABETES RISK

Preventive healthcare through predictive classification

This project used CDC BRFSS indicators to identify individuals at elevated risk for diabetes and support earlier screening, preventive outreach, and resource prioritization.

Cleaned and classified demographic and behavioral variables
Compared Logistic Regression and tree-based models
Prioritized recall, precision, F1, and ROC-AUC
Translated model results into prevention recommendations

Business questionWho should receive earlier preventive attention?

Primary riskFalse negatives in high-risk individuals

Decision useScreening and outreach prioritization

Portfolio lessonAccuracy alone is not enough in healthcare

PROJECT 03 · HIGH-COST PATIENTS

Predicting future cost before crisis utilization

This project used longitudinal healthcare expenditure data to identify patients likely to become high-cost users, supporting earlier intervention and more strategic resource allocation.

Used prior-year indicators to predict future high-cost status
Evaluated discrimination, ranking, and lift
Focused on operational use of the top-risk segment
Connected analytics to preventive financial strategy

ROC-AUC0.817 on sealed validation

Top-decile lift4.31× concentration of high-cost cases

Cases captured42.7% in the highest-risk decile

Portfolio lessonRanking quality can drive intervention strategy

PROJECT 04 · HOSPITAL OPTIMIZATION

Forecasting demand and prescribing resource decisions

This project combined CDC hospital operational data, machine learning forecasts, and a resource optimization model to recommend surge beds and additional nursing capacity.

Created lag, rolling, seasonal, and operational-stress features
Compared baseline, Linear Regression, Random Forest, and XGBoost
Used chronological train-test splits
Converted forecasts into staffing and bed recommendations

Selected modelXGBoost

Average R²0.907

Average RMSE2.888 occupancy points

Portfolio lessonPrediction is strongest when tied to action

Project viewing experience

Explore the work without downloading protected materials.

Each case study includes the business problem, methods, selected charts, performance results, and strategic recommendations. Full code, datasets, notebooks, workbooks, and reports remain confidential.

View Financial NLP ProjectSEC filing language, financial features, OLS, and time-aware validation. View Diabetes Risk ProjectClassification, prevention, feature importance, and risk segmentation. View High-Cost Patient ProjectFuture cost prediction, top-decile lift, and preventive intervention. View Hospital Optimization ProjectForecasting, model comparison, staffing, and surge-bed recommendations.

Interactive project previews

View the research, results, and business impact online.

Visitors can explore selected charts, methods, results, and recommendations directly on the website. Complete code, datasets, notebooks, workbooks, and reports remain confidential and cannot be downloaded.

Protected portfolio: These previews are for professional review only. Public downloading and redistribution are disabled.

OLS coefficients from the SEC filing text project

PROJECT 01 · FINANCIAL NLP

Stock Return Prediction Using SEC Filing Text

This study evaluated whether filing language adds predictive value beyond traditional financial metrics. The analysis compared financial-only, text-only, and combined models using time-aware validation.

MethodsOLS, NLP features, chronological splits

FocusIncremental value of textual information

ValidationMultiple train-test ratios and leakage buffer

Business valueTransparent investment evidence

What visitors can review

Research question and model design
Financial, textual, and combined feature strategy
Selected coefficient visualization
Interpretation of predictive limits and market efficiency

ROC curves from the diabetes risk project

PROJECT 02 · HEALTHCARE RISK

Early Diabetes Risk Prediction

Built classification models using public health indicators to identify individuals at elevated diabetes risk and support earlier screening and preventive outreach.

MethodsLogistic Regression, Random Forest, XGBoost

MetricsRecall, precision, F1, ROC-AUC

Risk priorityReducing false negatives

Business valueTargeted preventive intervention

What visitors can review

Model comparison and ROC analysis
Feature importance and risk segmentation
Healthcare interpretation and prevention strategy
Implementation and stakeholder considerations

PROJECT 03 · HEALTHCARE FINANCE

Future High-Cost Patient Prediction

Used longitudinal healthcare expenditure data to identify patients likely to become high-cost users before crisis care, enabling earlier preventive and financial intervention.

ROC-AUC0.817 on sealed validation

Top-decile lift4.31×

Cases captured42.7% in the top decile

Business valueEarlier intervention and cost control

What visitors can review

Longitudinal prediction design
Risk ranking and lift interpretation
High-risk segmentation strategy
Financial and preventive-care implications

PROJECT 04 · FORECASTING + OPTIMIZATION

Hospital Resource Optimization

Combined next-week occupancy forecasting with operational optimization to recommend staffing and surge-bed decisions before hospital demand becomes critical.

Selected modelXGBoost

Average R²0.907

Average RMSE2.888 occupancy points

Business valueForecast-driven resource allocation

What visitors can review

Forecasting pipeline and model comparison
Operational stress and temporal features
Optimization recommendations by demand scenario
Deployment, monitoring, and executive decision support