Datasets:

ADAMlam-16
/

coffee-data_eda-project

license: mit
task_categories:
  - tabular-classification
configs:
  - config_name: default
    data_files:
      - split: train
        path: coffee data.csv

Coffee Intake & General Health — Exploratory Data Analysis (EDA)

The Video

1. Objective

This project investigates whether a relationship exists between coffee intake levels and key health indicators — sleep, BMI, heart rate, and stress — using a 10,000-person synthetic health dataset sourced from Kaggle. The analysis follows a structured EDA pipeline: cleaning, transformation, segmentation, visualization, and correlation testing.

2. Dataset Overview

The dataset contains individual-level health and lifestyle measurements used to explore the relationship between coffee consumption and general health outcomes.

Raw size: 10,000 rows × 16 features
Clean size: ~9,400 rows × 15 features (after filtering + column removal)

Feature Categories

Feature	Type	Description
Age	Numeric	Age of the individual
Gender	Categorical	Male / Female
Country	Categorical	Country of origin
Coffee_Intake	Numeric	Daily coffee consumption (cups/day)
Caffeine_mg	Numeric	Estimated daily caffeine intake (mg)
Sleep_Hours	Numeric	Average nightly sleep duration
Sleep_Quality	Categorical	Self-reported sleep quality
BMI	Numeric	Body Mass Index
Heart_Rate	Numeric	Resting heart rate (BPM)
Stress_Level	Categorical	Self-reported stress level (Low / Medium / High)
Physical_Activity_Hours	Numeric	Weekly physical activity hours
Occupation	Categorical	Occupation category
Smoking	Numeric	Smoking indicator
Alcohol_Consumption	Numeric	Alcohol consumption level
Coffee_Drinker_Type	Categorical	Engineered group: Low / Moderate / High (quantile-based)

Target Variable

Coffee intake group — an engineered categorical variable created by splitting the continuous Coffee_Intake column into three quantile-based groups of roghly equal size:

Low — bottom third of daily intake
Moderate — middle third
High — top third

This variable is used throughout as the primary grouping factor for health comparison.

3. Methodology

3.1 Data Cleaning

Column Removal

The Health_Issues column was removed from the dataset. Upon review, it was found to be inconsistently populated and lacked a clear, reliable definition compared to the other health metrics available. Removing it preserved data integrity without meaningful information loss.

Invalid Value Filtering

Exploratory checks using df.describe() revealed that Sleep_Hours contained values as low as 3 hours per night. This was identified as implausible for a population-level health dataset and inconsistent with human physiology.

Decision: All rows where Sleep_Hours < 5 were removed.

before = len(df)
df = df[df['Sleep_Hours'] >= 5].reset_index(drop=True)
after = len(df)
print(f"Removed {before - after} rows with Sleep_Hours < 5")

This filtering was applied before re-engineering the Coffee_Drinker_Type groups to ensure the group boundaries reflected the cleaned population, not the raw one.

Duplicate Check

print(df.duplicated().sum())  # → 0

No duplicate rows were found.

3.2 Feature Engineering

Coffee_Drinker_Type was created using quantile-based binning (pd.qcut, q=3), which ensures equal sample sizes across groups and makes group comparisons statistically fair. Equal-width binning was evaluated but rejected because coffee intake is right-skewed — equal-width bins would have produced very uneven groups.

df['Coffee_Drinker_Type'] = pd.qcut(
    df['Coffee_Intake'], q=3, labels=['Low', 'Moderate', 'High']
)

3.3 Group Distribution

After cleaning and re-binning, the three groups are near-equal in size, confirming the quantile approach worked as intended.

4. Exploratory Data Analysis (EDA)

Research Question

Is there a measurable relationship between high coffee intake and general health outcomes?

Research Question 1

Does coffee intake level affect sleep duration?

Observation:

A clear downward trend is visible across groups: Low intake has the highest median sleep hours, High intake has the lowest.
The overlap between groups is present, indicating that coffee is one of several factors — not the only driver of sleep duration.

Correlation Analysis:

To move beyond group comparison and test whether the relationship holds continuously, a Pearson correlation and regression analysis was run directly on the raw Coffee_Intake variable:

Statistical Result:

Metric	Value
Pearson r	−0.17
Direction	Negative
Strength	Weak
p-value	< 0.001

Insight: The negative correlation (r = −0.17, p < 0.001) is statistically significant and consistent across all three intake groups. Higher coffee consumption is weakly associated with reduced sleep duration. The relationship is real but modest — coffee intake is one contributor among several factors affecting sleep.

Conclusion: Coffee intake has a weak but statistically significant negative association with sleep hours. It is the clearest directional signal in the dataset.

Research Question 2

Is BMI associated with coffee intake level?

Observation:

The boxplots show nearly identical medians and IQRs across all three groups — the distributions are visually almost indistinguishable.
Outliers appear equally across all three groups, consistent with a real-world population.
Unlike sleep, there is no visible directional shift in BMI from Low → High intake.

Insight: The three groups show virtually no difference in BMI distribution. This is a meaningful null finding — it suggests that coffee intake, at least in this dataset, does not meaningfully track with body weight. BMI is driven by many lifestyle variables that are not captured here.

Conclusion: BMI shows no meaningful association with coffee intake level. This variable does not carry a useful signal for the research question.

Research Question 3

Does coffee intake affect resting heart rate?

Observation:

The violin plots reveal that the distribution of resting heart rate shifts rightward as coffee intake increases.
The Low intake group is more tightly centered around a lower BPM, while the High intake group shows both a higher center and a wider spread.
The violin shape for the High group suggests greater variability — some individuals are unaffected while others show notable elevation.

Insight: The High intake group shows a measurable upward shift in resting heart rate, consistent with caffeine's known stimulant effect. The widening of the distribution in the High group is analytically interesting — it may reflect individual differences in caffeine sensitivity or confounding lifestyle variables.

Conclusion: Coffee intake is associated with modestly elevated resting heart rate. The effect is visible in distribution shape rather than median alone, which is why a violin plot was chosen over a boxplot for this metric.

Research Question 4

Does coffee intake influence stress level distribution?

Observation:

The bar chart shows that the vast majority of respondents in all groups report Low stress (70–82%), with Medium stress accounting for 18–29%.
High stress is reported by very few respondents overall: 0.5% (Low), 0.7% (Moderate), and 1.1% (High intake).
Despite the small absolute numbers, the directional pattern is consistent — as coffee intake increases, the Low-stress share shrinks (81.8% → 70.3%) and the High-stress share grows (0.5% → 1.1%).

Insight: The stress composition shift from Low → High intake is consistent and directional. Whether this reflects a causal relationship (caffeine increasing stress) or a selection effect (already-stressed individuals consuming more coffee) cannot be determined from observational data alone. Both are plausible mechanisms.

Conclusion: Coffee intake is associated with a higher proportion of high-stress reporters and a lower proportion of low-stress reporters. This is one of the most visually clear findings in the dataset.

Research Question 5

In a focused investigation into reports of high stress levels, are there any clear signs of a connection between group affiliation and age group on the number of people reporting high stress levels?(a deep dive into high stress reporters)

Observation:

Absolute counts of high-stress reporters: Low intake n=16 (0.5% of group), Moderate n=20 (0.7%), High n=31 (1.1%).
Although the absolute numbers are small, the High intake group has nearly double the rate of high-stress reporters compared to the Low group.
The age distributions across all three groups are broadly similar, with median ages in the low-to-mid 30s and comparable spreads.

Insight: The high-stress signal in the High intake group is not concentrated in a specific age bracket — it is distributed across the full age range. This makes it less likely to be a confound driven purely by age and more consistent with a genuine relationship between intake level and reported stress.

Conclusion: High-intake coffee drinkers are overrepresented among high-stress reporters across all age groups, strengthening the stress finding from Research Question 4.

Research Question 6

Within the high-stress group, is there a relationship between coffee intake and age?

Observation:

The scatter plot of coffee intake vs. age within the high-stress group shows a very slight positive trend.

Research Question 6

What has a bigger effect on high stress — age or coffee intake?

Observation:

The chart directly compares the absolute correlation strength of Age (r = 0.008) and Coffee Intake (r = 0.041) with high-stress reporting within the high-stress group.
Coffee Intake is roughly 5× stronger than Age as a predictor of high-stress status.
Both effects are small in absolute terms, but the comparison clearly identifies coffee intake as the dominant factor between the two.

Statistical Result:

Metric	Value
Age effect (absolute r)	0.008
Coffee Intake effect (absolute r)	0.041
Dominant factor	Coffee Intake (5× stronger than Age)

Insight: Age contributes virtually no signal (r = 0.008) while coffee intake, though still weak in absolute terms (r = 0.041), is meaningfully larger by comparison. This rules out age as a confounding explanation for the stress pattern observed across coffee groups.

Decision: Age is not a meaningful confound. The stress finding is attributed to coffee intake level, not to demographic age differences between groups.

Conclusion: Coffee intake is a stronger predictor of high-stress reporting than age within this group. Age can be safely excluded as an alternative explanation for the stress findings.

5. Correlation Structure

Variable	Correlation with Coffee_Intake	Direction
Sleep_Hours	−0.17	Negative (weak)
Heart_Rate	~+0.18	Positive (weak)
BMI	~+0.02	Negligible
Stress_Score	~+0.04	Negligible
Age	~+0.008	Negligible

Key finding: Sleep and heart rate are the two variables most associated with coffee intake in this dataset. BMI, stress score, and age show negligible linear correlations.

6. Final Conclusion

This analysis shows that high coffee intake is associated with a consistent directional pattern across health indicators: less sleep, modestly elevated heart rate, and a higher share of high-stress reporters. The effect is clearest for sleep (r = −0.17, p < 0.001) and heart rate, while BMI shows no meaningful association.

No single metric shows a dramatic signal, but the directional consistency across independent outcomes — sleep, heart rate, and stress all pointing the same way — is the most meaningful finding in the dataset.

Key Takeaways

Health Metric	Finding	Signal Strength
Sleep Hours	Negative association — High group sleeps less per night	Weak (r = −0.17)
Heart Rate	Positive association — High group shows elevated BPM	Weak
BMI	No meaningful difference across groups	Negligible
Stress Level	High group has highest share of high-stress reporters	Small but consistent
Age	No meaningful relationship with intake level	Negligible

7. Limitations

The dataset is synthetic — relationships are plausible but may not reflect real-world causal mechanisms.
The analysis is observational — no causal claims can be made. Directionality (does coffee cause poor sleep, or do poor sleepers drink more coffee?) cannot be determined here.
Coffee_Drinker_Type groups are constructed from quantile splits; different binning schemes may shift group boundaries slightly.
Self-reported variables (Stress_Level, Sleep_Quality) are subject to measurement noise.
Confounders such as occupation, smoking, and alcohol consumption were not controlled for in the univariate analyses presented here.

8. Notebook & Plots

Full analysis with code: Google Colab Notebook