Creative Projects
08 Creative Projects: Turning the Police Use‑of‑Force Dataset into a Flagship Data Science Portfolio
Zara Okonkwo, for Data Science and Statistics programs at Berkeley, Carnegie Mellon, and Georgia Tech, the most convincing evidence of readiness is a technically rigorous, publicly visible data project. The committee flagged the police use‑of‑force dataset you referenced as the strongest foundation for such a project. Right now, it likely reads as a data analysis or research activity. Your opportunity before application deadlines is to elevate it into a full, professional-grade data science artifact: an open dataset, reproducible pipeline, and technical analysis that demonstrates both statistical depth and software engineering maturity.
The goal is not simply to show that you analyzed data. The goal is to demonstrate how you think like a data scientist: how you collect, clean, model, document, and publish data so that other people can use it.
Project Vision: An Open Policing Data Platform
Instead of presenting a one-off analysis, structure the project as a reusable research resource. Admissions readers in technical programs respond strongly to projects that combine statistics, software engineering, and public transparency.
Your project can be framed as an open policing analytics platform that includes three layers:
- Dataset: A clean, structured, downloadable dataset documenting police use-of-force incidents.
- Reproducible analysis: Statistical modeling exploring patterns and disparities.
- Public tools: An API or interactive dashboard allowing others to explore the data.
This structure signals exactly the type of thinking expected in modern data science programs.
Technical Architecture
To make the project credible at the level expected by Berkeley, CMU, and Georgia Tech, organize the repository like a professional data science project rather than a school assignment.
| Component | Purpose | Suggested Stack |
|---|---|---|
| Data Collection Pipeline | Automates acquisition and cleaning of use‑of‑force data | Python, Pandas, Jupyter, possibly BeautifulSoup if scraping |
| Data Processing | Standardizes fields, handles missing values, and documents assumptions | Pandas, NumPy |
| Statistical Modeling | Analyzes trends and potential disparities | Python (statsmodels, scikit‑learn) or R |
| Interactive Dashboard | Allows users to explore patterns visually | Plotly Dash, Streamlit, or Observable |
| Public API | Allows external researchers to query the dataset | FastAPI or Flask |
| Documentation | Explains methodology and limitations | Markdown documentation and technical report PDF |
The key signal here is reproducibility. Someone downloading your repository should be able to run the pipeline and recreate the analysis.
Advanced Statistical Modeling
Right now, the project likely focuses on descriptive analysis. To strengthen its academic credibility for Data Science and Statistics programs, extend the analysis with more sophisticated modeling.
Consider adding:
- Regression analysis examining relationships between incident variables (for example, location characteristics or time trends) and the likelihood of use‑of‑force events.
- Trend modeling to evaluate changes over time across jurisdictions or departments.
- Fairness analysis exploring potential disparities across demographic groups where data allows.
- Causal inference approaches (even exploratory ones) discussing whether policy changes appear associated with measurable shifts in outcomes.
Equally important is documenting limitations. Admissions readers appreciate students who acknowledge uncertainty, missing data, and bias in datasets. Transparency strengthens the credibility of the project.
The Technical Report
Your project should include a formal technical report that reads like a condensed academic paper. This report can be linked in your application and hosted in the GitHub repository.
Structure the report with sections such as:
- Introduction and motivation
- Data sources and collection methodology
- Data cleaning and preprocessing decisions
- Statistical models used
- Key findings and visualizations
- Limitations and ethical considerations
- Future work
Keep the tone analytical rather than advocacy-oriented. Programs like CMU and Berkeley especially respond to students who treat social datasets with methodological rigor.
Making the Project Reusable
A major step that distinguishes standout portfolios is enabling other people to build on the work.
Consider incorporating at least one of the following:
- Downloadable dataset with clear schema documentation.
- Public API allowing queries by year, region, or incident type.
- Interactive dashboard where users can filter and visualize trends.
This moves the project from “analysis” to “infrastructure.” That shift is extremely valuable for Data Science admissions.
GitHub Portfolio Strategy
Your GitHub repository should be structured like a professional open-source project. Many admissions readers—especially in technical departments—will glance at repository organization.
Suggested structure:
- /data – processed datasets
- /notebooks – exploratory analysis
- /pipeline – scripts for cleaning and transformation
- /models – regression or modeling code
- /dashboard – visualization application
- /docs – methodology and technical report
The README file should clearly explain:
- What the project does
- Why the dataset matters
- How to reproduce the analysis
- How to use the dashboard or API
A clear README alone dramatically improves how admissions readers perceive the project.
Demonstrating Real‑World Impact
If possible before application submission, track whether anyone actually uses the project. Even small signals of adoption strengthen credibility.
Examples include:
- GitHub stars or forks
- Researchers or students downloading the dataset
- Users interacting with the dashboard
- Mentions in community forums or data science communities
You do not need large numbers. Even limited external use demonstrates that the project has value beyond a class assignment.
How This Strengthens Your Application
For Data Science and Statistics admissions, this project demonstrates several qualities simultaneously:
- Statistical reasoning
- Data engineering ability
- Software development skills
- Ethical awareness around real-world datasets
- Ability to publish and communicate technical work
A well-executed version of this project can function as the centerpiece of your technical portfolio.
Application‑Season Build Calendar
| Month | Key Actions | Target Outcome |
|---|---|---|
| August |
|
Fully reproducible dataset and project framework |
| September |
|
Complete statistical analysis and early report draft |
| October |
|
Public-facing project ready for submission |
| November |
|
Fully polished portfolio artifact |
Zara Okonkwo, this project does not require inventing something entirely new. The leverage comes from elevating the work you already began into a rigorous, open, technically documented data science platform. Done well, it becomes a concrete demonstration of the type of analytical and engineering thinking programs like Berkeley, CMU, and Georgia Tech expect from future data scientists.