Implementation notes · Machine learning · Thesis support
Financial Aid Prediction & Analytics
Engineering and analytical contributions to a financial-aid graduate thesis authored by another student. My work covered cleaning operational data, validating features, developing eligibility and award-percentage modeling workflows, and exploring how their outputs could support human decisions.
- Python
- pandas
- scikit-learn
- Jupyter
Prepare
Parse, validate, and check data quality
Predict
Eligibility and award-percentage estimates
Review
Thresholds, uncertainty, and model behavior
I built data and modeling workflows in support of another student's graduate thesis.
The public repository is a sanitized notebook archive. Private records, trained models, and generated results are excluded; missing source modules prevent a complete end-to-end rerun from this archive alone.
Preparing data for analysis and modeling
The cleaning workflow parses operational spreadsheet exports, records quality flags, and separates raw, inferred, audit-only, and model-safe fields. This keeps data issues visible and helps prevent award outcomes and other post-decision information from becoming predictive inputs.
Pipeline architectureTwo prediction tasks, with review built in
Separate notebooks address eligibility classification and aid-percentage estimation. The eligibility workflow compares models, examines calibration and thresholds, and explores review bands for uncertain cases. The archive also documents error analysis, subgroup checks, and interpretation of model behavior.
Eligibility modeling workflowEstimating the award percentage
The percentage notebook documents the regression workflow alongside prediction diagnostics and decision-support analysis. It complements the eligibility question by examining the level of aid awarded. The public snapshot supports review of the approach, rather than independent verification of unpublished model results.
Aid-percentage modeling workflow