Individual Retrospective — Milestone 3: Data Analysis

[salihakalender]

Hello,

Please share your individual learning retrospectives for this Milestone 3: Data Analysis

🧠 What you learned

✅ What went well

⚠️ What could be improved

[SiSaR-Pal]

🧠 What I learned

• Model Building: The Universal AI modules and recitations were invaluable for understanding how to properly build, test, and interpret models. I learned that a high R-squared score can be a red flag for overfitting, and a more generalizable model with a lower score is often superior.
• Insightful EDA: Stratifying exploratory data analysis by development level (SDI) from the beginning is critical. Global averages can mask the most important underlying trends in diverse datasets.
• Explanatory Power: Replacing hundreds of country-specific dummy variables with a single, powerful explanatory variable like SDI leads to a more robust and scientifically valid model.

✅ What went well

• Literature Review: The foundational knowledge gained from the WHO, IDMI, and A State of Global Air reports was instrumental in guiding the feature engineering and analysis phases.
• Modular Workflow: Breaking the analysis into five focused notebooks created a clean, reproducible pipeline that was easy to debug and manage.
• Iterative Refinement: The step-by-step process of developing the analysis allowed for immediate feedback and course correction, which significantly improved the final outcome.

⚠️ What could be improved

• Initial Visualizations: My first attempts at creating time-series plots were not insightful. I need to start with stratified or faceted plots in the future to get to the core story faster.
• Local Automation: I could improve my workflow by using local pre-commit hooks to automatically format code. This would prevent CI/CD pipeline failures and streamline the development process.
• Advanced Modeling: While Random Forest provided a strong baseline, I could explore more advanced techniques like Mixed-Effects Models in the future to better account for the data's hierarchical structure.

[Mohamedmxz]

Mohamed Tilal — Milestone 3 Retrospective

What I Learned
I learned new ML models and applied some modeling techniques like Linear Regression and Random Forest, and I practiced doing exploratory data analysis (EDA) independently. This helped me better understand the connection between data and insights.

What Went Well
I contributed to the data analysis, helped in reviewing the old dataset and collecting new ones, and reviewed reports to support our modeling decisions. The team was mostly active and supportive, which made collaboration smooth and productive.

What Could Be Improved
Applying the models and understanding their outputs was a bit challenging at first, but it became easier with practice. Also, a few team members were inactive, which slightly affected the overall team flow, I know some of them have difficult circumstances and I am totaly ok with it just let the team know and stay following the process.

[salihakalender]

🧠 What I Learned

• Thanks to the Universal AI modules and recitations, I was able to better understand and apply machine learning analyses. This process was highly valuable for me.
• Through the feedback provided by my teammates, I learned how to improve analyses.

✅ What Went Well

• After team decisions were made, communication became more fluent, which helped tasks proceed quickly and smoothly.
• Working iteratively and receiving feedback at each step improved the quality of the analysis.
• The strategic decisions we made regarding the data and model strengthened the scientific soundness of the project.

⚠️ What Could Be Improved

• I should deepen my knowledge and practice of advanced modeling techniques.
• Gaining more experience in model optimization and parameter tuning would be beneficial.
• Clarifying roles and task distributions earlier would help improve workload management and increase overall efficiency.

[Adamx090]

Salih Adam

🧠 What you learned

• Understanding the dataset and its measurements is critical:
  Deeply exploring what each metric means, how it is measured, and how it relates to real-world phenomena is essential for drawing valid conclusions in data analysis.

• Practiced computing and interpreting statistical relationships:
  Learned how to calculate and interpret correlation coefficients and p-values, and understood their role in identifying potential associations between variables.

• Recognized the importance of temporal dynamics in data analysis:
  Became aware that outcomes in many datasets may not occur simultaneously with their causes or exposures, and that considering time lags is crucial when analyzing trends over time.

✅ What went well

• Clear trend visualizations and correlations computed:
  The analysis revealed meaningful insights about how PM₂.₅ levels relate to health burdens, particularly for cardiovascular and respiratory outcomes.
• Reproducible pipeline built:
  The class-based approach allowed for systematic data handling, making extraction and analysis flexible and scalable.

⚠️ What could be improved

• Gain deeper insight into data analysis by completing the Universal AI module:
  Advancing knowledge in machine learning, modeling, and advanced analytics will help extend this work beyond basic correlations.

• Improve team coordination and communication:
  Streamlining how the team shares progress, discusses findings, and aligns on next steps can enhance efficiency and collaboration.

[ObayCipher]

🧠 What You Learned

I learned that understanding the type of dataset and its structure is a key starting point in any data analysis process. It allows you to identify patterns and relationships between variables more effectively. I also recognized how statistical parameters such as means, correlations, and variability are essential tools in exploring and interpreting data, especially when working with environmental and health-related indicators.

✅ What Went Well

One of the successes during this milestone was being able to translate results into clear visualizations. This helped in making connections between different variables more intuitive and accessible. It was rewarding to see how data trends can be communicated visually to tell a compelling story.

⚠️ What Could Be Improved

I would like to improve my ability to work with different types of data, including structured and unstructured datasets. Understanding how to clean, manipulate, and analyze various data formats is something I plan to continue learning. I also aim to dive deeper into the Universal AI (UAI) course to gain more knowledge about advanced data analysis techniques and build more confidence in handling complex datasets.

[linahKhayri]

Linah Khayri

🧠 What I Learned

• Multifactorial health outcomes require contextual modeling
  We learned that analyzing national COVID-19 mortality involves not only environmental exposures like PM₂.₅, but also socio-demographic factors and a lot of confounding variables.

• MLR and ML each offer distinct value
  Traditional regression models gave us interpretable but modest associations, while machine learning methods revealed more complex patterns that were less transparent but potentially more powerful.

• Accounting for age-standardized rates made a big difference
  Adjusting for age-standardized rates strengthened the results and helped with comparisons across countries.

• Model performance vs. interpretability
  Balancing interpretability and predictive performance is tricky, especially in public health research where clear communication is key.

✅ What Went Well

• Well-documented code and structured notebooks
  Jupyter notebooks were organized and well-commented, which made it easier to revisit and build on models.

• Diverse modeling approaches
  We used a mix of statistical methods (MLR, interaction terms, spline regression) and machine learning (random forest, quantile regression), which gave us a broader view of the data.

⚠️ What Could Be Improved

• More robust feature selection
  We could have added more variables like comorbidities, population density, healthcare capacity, or mobility data to better account for confounding.

• Improve team collaboration
  Team coordination was okay, but there’s room to improve communication and workflow, especially in early planning stages

[FalaqMajeed]

What I Learned
Health outcomes are shaped by multiple, interacting factors
We learned that analyzing global COVID-19 mortality requires more than just looking at PM₂.₅ levels—socio-demographic factors, health disparities, and confounding variables must be considered together.

Each modeling approach reveals different layers
Spline regression helped capture non-linear relationships between PM₂.₅ and mortality, while random forest modeling uncovered deeper, more complex interactions that weren’t visible in traditional methods.

Standardizing by age matters
Using age-standardized death rates allowed us to make fairer comparisons across countries with different population structures and improved the reliability of our findings.

Interpretability vs. complexity is a trade-off
Spline regressions and interaction models were more transparent, while random forests offered stronger predictive insights but less clarity in interpretation—highlighting a key tension in health data science.

✅ What Went Well
Clear, well-structured notebooks
Our Jupyter notebooks were clean, organized, and well-annotated—making it easier to troubleshoot, share, and revisit models.

Diverse modeling techniques
We applied a strong mix of analytical tools—spline regression, interaction terms, and machine learning (random forest, quantile regression)—which gave us richer, more nuanced insights.

⚠️ What Could Be Improved
More comprehensive variable selection
We could enhance our models by adding variables like comorbidity prevalence, healthcare system capacity, or population density to better account for known confounders.

Earlier alignment on team planning
Our collaboration was effective overall, but earlier alignment on goals, roles, and timelines could’ve prevented delays and made the workflow smoother.