1 Reliability (Data reliability)

The reliability of the Fitabase/Bellabeat data is overall moderate, with several limitations.

1.1 Strengths

  • Data collected from real devices (Fitbit), therefore objective measurements: no self-reporting.
  • Consistent timestamping across the minute-by-minute, hourly and daily files.
  • No critical missing values in time columns and user identifiers.
  • Rich granularity: minute, hour, day → suitable for time-series analysis and pattern detection.

1.2 Weaknesses

  • Very small sample: only 30 users, which strongly limits generalizability.
  • Uneven distribution of user contributions: some users provide a lot of heart-rate data, others almost none (e.g. heartrate_seconds_merged).
  • Imbalance in some metrics:
    • minuteMETsNarrow_merged: extremely skewed distribution;
    • weightLogInfo_merged: very low coverage → potential bias.
  • Data collected over a short period (31 days) → no annual seasonality.

Reliability conclusion – Sufficient for an educational exploratory project, not strong enough to support robust market research recommendations.

2 Originality (Uniqueness / analytical value)

2.1 What the data allows us to analyse

  • Circadian patterns thanks to the minute- and hour-level files.
  • Global behavioural analysis: activity, sleep, calories, heart rate.
  • Multi-granularity combination → rare and valuable for modelling a typical day.
  • Possibility to reconstruct a complete user journey:
    • sleep → wake-up;
    • activity / intensity → calories → METs → daily behaviour.

2.2 Limitations

  • No socio-demographic variables → no profile-based analyses (age, gender, generalizable BMI, etc.).

Originality conclusion – The diversity of detail levels is the main strength of this dataset.

3 Comprehensiveness (Coverage / functional scope)

3.1 Analytical coverage provided by the 11 files

Area Coverage level Comment
Daily activity High dailyActivity, dailyIntensities, dailySteps → complete global view.
Hourly activity Very high hourlyCalories, hourlyIntensities, hourlySteps → robust circadian analyses.
Minute-level activity Very high Fine granularity for modelling or detecting activity peaks.
Calories / energy expenditure High minuteCaloriesNarrow + hourlyCalories → consistent measurement over time.
METs (physiological intensity) High Rare metric but highly skewed.
Sleep Moderate minuteSleep_merged → good level of detail but no sleep stages.
Heart rate Low heartrate_seconds_merged incomplete depending on the user.
Weight / BMI Very low weightLogInfo_merged almost unusable for global analyses.

3.2 Temporal comprehensiveness

  • 31 days → sufficient for:
    • daily patterns;
    • behavioural clustering;
    • habit quantification.
  • Insufficient for:
    • seasonality;
    • long-term behaviour change.

Comprehensiveness conclusion – High for activity, moderate for sleep, low for heart rate and weight/BMI.

4 Citation (Documentation / traceability / reproducibility)

4.1 Strengths

  • Clearly named files.
  • Homogeneous columns across files (Id, dateTime, value).
  • Fitabase documentation is publicly available.

4.2 Weaknesses

  • No metadata embedded in the files.
  • No device identifiers → loss of contextual information.
  • No complete official README.

Citation conclusion – Intrinsically weak at the raw file level; improved only through project documentation.

5 Currency (Data recency)

The data dates back to 2016.

5.1 Consequences

  • 2016 Fitbit devices → significant technological bias.
  • Health guidelines, activity recommendations and intensity classifications have evolved since then.
  • User behaviour has changed (more smartphone integration, more modern sensors).

Currency conclusion – Weak for operational decision-making, but adequate for an academic or training-oriented analytics project.

Final ROCCC Summary

5.2 Strengths

  • Exceptionally rich granularity (minute → hour → day).
  • Temporal consistency.
  • Real, non self-reported data.
  • High potential for behavioural analysis and segmentation.
  • Dataset well-suited for practising EDA, cleaning, profiling, clustering and circadian analysis.

5.3 Weaknesses

  • Sample size too small (30 users → low statistical reliability).
  • Old data (2016).
  • Strongly skewed distributions for METs, intensity and minute-level calories.
  • No demographic variables.

5.4 Overall conclusion

The 11 Bellabeat files provide an excellent learning ground for data analytics: profiling, EDA, quality checks, ETL integration, visualisation, segmentation, and building an analytical narrative.

However, they are too limited for real-world decision-making, mainly because of:

  • the small sample size;
  • the lack of user diversity;
  • the age of the data.