Global Coffee Health Dataset
Using exploratory data analysis, statistical testing, and machine learning, our team analyzed a global dataset to investigate whether coffee consumption is associated with stress and other health outcomes.
01 · Overview
Coffee is one of the world's most consumed beverages, influencing daily habits, productivity, and health. Because caffeine affects sleep, energy, and stress, understanding these relationships can help individuals, healthcare professionals, and the coffee industry make more informed decisions.
02 · Problem statement
Many people believe that drinking more coffee causes higher stress. However, evidence is limited on whether coffee itself is responsible or whether factors such as sleep quality and health play a much larger role.
Research question
Does coffee consumption have a statistically significant relationship with stress and other health indicators?
Project goal
We brought together exploratory data analysis, statistical testing, regression, machine learning, and data visualization to test one central claim.
03 · Dataset
Source: Kaggle's Global Coffee Health Dataset. The dataset captures coffee habits, health, sleep, BMI, and lifestyle variables across a broad international sample.
04 · My contributions
Prepared, transformed, and standardized the dataset for analysis.
Identified trends through summary statistics and visualizations.
Applied hypothesis testing and regression models.
Built and evaluated an XGBoost classifier.
Created visualizations with Seaborn and Matplotlib.
Interpreted results and communicated findings with teammates.
05 · Process
Reviewed literature on coffee consumption and health outcomes.
Filled missing values, converted categories, created a cups-of-coffee feature, and standardized data types.
Used descriptive statistics, a correlation matrix, boxplots, and distribution analysis.
Applied a Chi-Square test, Cramér's V, and OLS regression.
Evaluated XGBoost with cross-validation, a confusion matrix, a classification report, and feature importance.
Designed charts that made comparisons and model evidence understandable.
Synthesized evidence into practical findings and recommendations.
06 · Tools
07 · Key visualizations
Compared coffee consumption and reported health issues across 20 countries to identify broad geographic trends.
Sleep duration decreased consistently as stress increased, suggesting a much stronger relationship than coffee consumption.
Machine learning confirmed that sleep, age, and heart rate contributed far more to stress prediction than coffee consumption.
08 · Key findings
The XGBoost model achieved about 82% accuracy, while coffee ranked among the least important predictors compared with sleep-related variables.
09 · Challenges
More than 5,900 Health Issues values required imputation.
Converted categorical variables into usable numerical features.
Compared multiple approaches before selecting XGBoost.
Separated statistical significance from practical significance.
The report also discusses synthetic data, missingness, feature engineering decisions, and modeling tradeoffs as important limitations.
10 · Final insight
Although coffee consumption showed a statistically detectable association with stress, its practical effect was very small.
Sleep duration, sleep quality, health conditions, and age consistently explained stress levels much better than coffee intake.
11 · Skills demonstrated
Python · Pandas · Data Cleaning · EDA · Statistical Analysis · Hypothesis Testing · OLS Regression · XGBoost · SHAP · Seaborn · Matplotlib
Team Collaboration · Data Storytelling · Critical Thinking · Research · Analytical Communication · Problem Solving
12 · Reflection
This project taught me to distinguish statistical significance from practical significance.
I strengthened my ability to move beyond exploratory analysis and use statistical testing and machine learning to answer a real research question. I gained experience cleaning complex data, evaluating predictive models, interpreting results, and communicating evidence through clear visualizations.
Explore on GitHub