Retrospective— Milestone 3: Data Analysis

[salihakalender]

Hello team,
Please share your thoughts under each section below, and add your own individual retrospective at the bottom.

Retrospective — Milestone 3: Data Analysis

✅ What Went Well

⚠️ What Was Challenging

💡 What We Learned

🔄 What We’ll Improve Next

🎯 Strategy vs. Board

✅ What parts of your plan went as expected?

⚠️ What parts of your plan did not work out?

➕ Did you need to add things that weren't in your strategy?

➖ Or remove extra steps?

Individual Retrospectives

[SiSaR-Pal]

Retrospective — Milestone 3: Data Analysis

✅ What Went Well

• Modular Workflow: Splitting the analysis into five focused notebooks was highly effective for clarity and reproducibility.
• Iterative Refinement: The step-by-step development process allowed for quick feedback and immediate improvements, leading to a stronger final analysis.
• Key Insight Pivot: Abandoning a flawed, overfitted model in favor of one using the Socio-Demographic Index (SDI) was a crucial success. It produced a more scientifically valid and interpretable result.

⚠️ What Was Challenging

• Initial Visualizations: Early global-average plots were misleading. It took several iterations to develop faceted visualizations that revealed the true, underlying trends.
• Explaining Model Performance: The most significant challenge was communicating why a lower R-squared value represented a better, more realistic model after moving away from an overfitted one.
• CI/CD Integration: Resolving the initial code formatting errors flagged by the Ruff linter required a few cycles, presenting a brief learning curve with the project's quality assurance tools.

💡 What We Learned

• High R² as a Red Flag: An almost-perfect R-squared score is often a symptom of severe overfitting, and a model with a lower score that generalizes better is more valuable.
• Stratify Early, Stratify Often: For diverse global datasets, stratifying the exploratory analysis (e.g., by development level) from the start is essential to uncover meaningful insights.
• The Value of Automation: Automated code formatting tools like Ruff are invaluable for maintaining project consistency and professional standards.

🔄 What We’ll Improve Next

• Start with Stratified EDA: Begin all future analyses with stratified visualizations to get to the core insights faster.
• Local Pre-Commit Hooks: Implement local hooks to run formatters automatically before committing code, preventing CI pipeline failures.
• Explore Advanced Models: Build on the current Random Forest baseline by exploring more advanced techniques like Mixed-Effects Models to better handle the data's structure.

🎯 Strategy vs. Board Summary

• ✅ As Planned: The overall phased approach (Data Prep -> EDA -> Modeling -> Interpretation) and the feature engineering were executed successfully.
• ⚠️ Unplanned Pivot: The initial modeling strategy from the old project was fundamentally flawed and had to be completely replaced, not just adjusted.
• ➕ Added Steps: We added several revised visualization steps in response to feedback and a new analysis to explain the R-squared drop. Debugging the CI formatting issues was also an unplanned but necessary task.
• ➖ Replaced Steps: The initial modeling methodology was completely swapped out for a more robust approach using SDI as the primary explanatory variable.

[Mohamedmxz]

Team Retrospective:

✅ What Went Well
The team created a clear plan and followed it effectively. Every member was helpful and supportive whenever needed. We explored various datasets and tried multiple analysis approaches to strengthen our research and improve our model.

⚠️ What Was Challenging
It was difficult to recognize issues in the data and fix them. We also found it challenging to understand and apply different analysis and modeling techniques, especially when interpreting the results in a meaningful way.

💡 What We Learned
You can apply many types of analysis and build multiple models, but what really matters is how well your work answers the core research question. Keeping the objective in focus is key.

🔄 What We’ll Improve Next
Overall, the process went smoothly, but we realized that setting internal deadlines earlier would reduce stress and give us more time to reflect and improve before submission.

🎯 Strategy vs. Board Reflection
✅ What parts of your plan went as expected?
We followed our plan closely, especially in terms of dividing tasks, exploring different analysis approaches, and making team decisions collaboratively.

⚠️ What parts of your plan did not work out?
We underestimated how much time it would take to understand data issues and apply the right modeling methods. Some steps took longer than expected due to technical or conceptual challenges.

➕ Did you need to add things that weren't in your strategy?
Yes — we had to explore additional datasets and test extra analysis methods to find more reliable insights and strengthen our conclusions.