DATA ANALYSIS & INTEGRATION
I'm Benjamin Buckland, a data analyst with experience preparing and validating biomedical data in Python and SQL. I combine technical data work with a background in operations, stakeholder coordination, and clear documentation.
Python · SQL · R · Biomedical data · Knowledge graphs
THE WAY I WORK
Good analysis starts before the model.
Define the question. Understand the data. Make the limitations clear.
01 / SELECTED WORK
A closer look at the process
behind the results.
FEATURED CASE STUDY · PUBLIC RETAIL DATA
A customer retention analysis that connects Python preprocessing, Snowflake SQL, and a simple predictive model.
1,605 of 3,370 eligible customers returned.
Measured from each customer's first observed purchase, not their first-ever purchase.
THE QUESTION
Measure how often customers purchase again within 90 days, then test whether their first order helps predict that outcome.
Dataset: Chen, D. (2015), UCI Online Retail, December 2010 to December 2011. Data and derived results: CC BY 4.0.
THE WORKFLOW
THE RESULT THAT MATTERS
The model only slightly outperformed the baseline. This logistic regression, using three first-order features, showed weak discrimination on the held-out test set. This version is not deployment-ready.
LOCAL PYTHON REPRODUCTION
Stratified 80/20 split, seed 42: 2,696 training and 674 test customers. Scaling fitted on training data only. Results describe this test set, not future production performance.
The figures shown here come from the committed local Python reproduction. The repository includes the Snowflake SQL workflow, but Snowflake SQL was not executed in this run.
Customers are sorted by customer ID and invoice number before the stratified split. The original showcase’s row order was not retained, so this is a reproducible evaluation rather than an exact replay of that earlier test set.
Each customer’s first observed order starts their own 90-day window. Customers are included only when the dataset continues for at least 90 days after that order.
Customers who do not return remain in the denominator. Customers without enough follow-up are excluded rather than labeled as non-returners. “First observed” is important because earlier purchases may fall outside this dataset.
Of 321 actual returners in the test set, the model identified 58. That is about 18.1% recall at a classification threshold of 0.5.
Accuracy alone hides this weakness. A useful next experiment would test additional information available at prediction time and use a time-based evaluation with fully observed outcomes.
These are proposed next steps, not completed improvements.
| Actual outcome | Predicted no return | Predicted return |
|---|---|---|
| No return | 302 | 51 |
| Return | 263 | 58 |
Return means another purchase within 90 days.
The observed repeat-purchase rate provides a reference for discussing retention. Repeating the same definition across comparable cohorts could help a business investigate whether customer behavior is changing.
The model is an exploratory test, not a customer-targeting recommendation. This project does not establish marketing uplift, causation, or how well the model would perform on future customers.
PROFESSIONAL EXPERIENCE · PREVIOUS ROLE
–
In my previous role as a Data Analyst at Vascentis, I worked on biomedical data integration, metadata organization, and ontology mapping.
A high-level description of professional experience. No proprietary datasets or internal materials are shown.
02 / BACKGROUND
My background spans program coordination, international project research, and biomedical data integration. Across that work, I have gathered requirements, organized complex information, and documented workflows so other people can use them.
I am continuing that path through a B.S. in Analytics at Northeastern University, alongside graduate coursework in Applied AI through the PlusOne program.
Vascentis · Data Analyst
–
Python, SQL, and PostgreSQL data mapping and validation; workflow refactoring with test datasets; and 39 pipeline visualizations.
See my Vascentis contributions →Northeastern University
Expected completion: May 2028. Degree in progress.
PlusOne program: graduate coursework toward an MPS in Applied AI.
Catena Resources · Remote
Researched community development opportunities across seven countries, comparing costs, operational requirements, and implementation risks. Synthesized stakeholder input into recommendations for leadership.
Catena Resources · Bolu, Türkiye
Gathered stakeholder requirements, translated needs into operational workflows, and tracked capacity, demand, and service quality.
03 / HOW I CAN CONTRIBUTE
Data cleaning, SQL aggregation, metadata structure, and reproducible analysis workflows.
Exploratory analysis, ontology mapping, and making relationships across data sources understandable.
Baseline comparisons, readable visualizations, and clear communication of results and limitations.
LET'S BUILD SOMETHING USEFUL
I'm interested in data analyst opportunities and focused project work in data preparation, analysis, and integration.