DATA3888: Data Science Capstone
A capstone project unit focused on building a public data product through interdisciplinary collaboration, computational analysis, and clear communication.
Learning Outcomes
- Problem framing: Study an interdisciplinary data-driven problem and formulate it into an analytical framework.
- Interdisciplinary application: Apply disciplinary knowledge to solve problems in an interdisciplinary context.
- Investigation strategy: Create an investigation strategy, explore solutions, discuss approaches, and predict outcomes.
- Digital analysis: Analyse data using modern information technology and digital skills.
- Resilience and integrity: Demonstrate integrity, confidence, personal resilience, and the capacity to manage challenges individually and in teams.
- Collaboration: Collaborate with diverse groups and across cultural boundaries to develop solutions to project problems.
- Communication: Communicate project outcomes effectively to an interdisciplinary audience.
Takeaways
The capstone flipped the usual unit structure: instead of learning a method and then applying it to a toy dataset, we started with a real public-interest problem and had to figure out which methods were actually appropriate. That reverse pressure — from question to tool, not tool to question — is how real data science work actually operates, and it is a harder skill to develop. Working in an interdisciplinary team across majors meant we constantly had to translate between technical and non-technical language. I learned that the gap isn't about intelligence; it's about shared vocabulary. Building that shared vocabulary early in a project saves enormous amounts of rework later. Communicating uncertainty honestly to a non-specialist audience was the hardest single skill. Stakeholders want a confident answer, but good data science often produces a range of plausible answers with caveats. Learning to present that honestly without losing the audience's trust was something I worked on throughout the entire semester. The biggest carry-forward is probably process discipline: scoping, versioning, documentation, and reproducibility matter just as much as the analysis itself when the output needs to be a product that others can use, audit, or build on.