Bellabeat Tech Case Study:

Table of Contents

• Introduction
  
• Step 1. Ask
  
  • 1.1 Ask: Identifying stakeholders and their expectations
    
  • 1.2 Ask: Statement of business task
    
• Step 2. Prepare
  
  • 2.1 Prepare: Dataset introduction
    
  • 2.2 Prepare: Does the data ROCCC?
    
  • 2.3 Prepare: Dataset limitations & data selection
    
• Step 3. Process
  
  • 3.1 Process: Examining the datasets
    
  • 3.2 Process: Examining the datasets
    
  • 3.3. Process: Data cleaning and verification
    
  • 3.4 The data cleaning and verification details
    
  • 3.5 Data merging/splitting possibilities
    
• Step 4. Analyze
  
  • 4.1 Analyze: General statistics
    
  • 4.2 Analyze: Identifying trends and relationships
    
• Step 5 Share
  
• Step 6 Act
  • 6.1 Act: Insights
    
  • 6.2 Act: Recommendations
    
• Skills Used
  
• Limitations and future scope
  
• References
  

Introduction

(Back to table of contents)

Bellabeat is a high-tech company that manufactures health-focused smart products. Sršen used her background as an artist to develop beautifully designed technology that informs and inspires women around the world. Collecting data on activity, sleep, stress, and reproductive health has allowed Bellabeat to empower women with knowledge about their own health and habits.

About Bellabeat

• Founders: Urška Sršen and Sando Mur

• Founded: 2013

• Core business: Manufacturer of health-focused smart products

• Mission statement: Empowering women to reconnect with themselves, unleash their inner strengths and be what they were meant to be

• Products:

  • Bellabeat app
  • Leaf (classic wellness tracker)
  • Time (wellness watch)
  • Spring (smart water bottle)
  • Bellabeat membership (membership program for users)

• Growth By 2016, Bellabeat had opened offices worldwide. Through extensive investments in advertising (traditional and digital, including Google Search and social media), Bellabeat products became available through an increasing number of online retailers.

In this case study, I'm going to define a business task and follow a data analysis approach in order to suggest new growth opportunities for the company

Bellabeat founders believe that there are even more growth opportunities out there, and an analysis of available consumer data would reveal them. The marketing analytics team has to make it happen by following the steps of the Google data analysis process: ask, prepare, process, analyze, share, and act

The primary objectives of this analysis are:

1. To identify trends in smart device usage, particularly physical activity and sleep behavior.
2. To determine how these trends can be applied to Bellabeat’s customer base.
3. To inform high-level marketing strategy by presenting data-driven recommendations.

The analysis follows the six phases of the data analysis process: Ask, Prepare, Process, Analyze, Share, and Act.

Using tools such as R and ggplot2, the study explores relationships between :

• total steps
• sedentary behavior,
• activity intensities,
• sleep efficiency

You can find the R script here

This offers insights into how daily habits may affect wellness outcomes.

Ultimately, this case study not only supports Bellabeat’s mission to empower women through health-focused technology, but also serves as a portfolio-ready project that showcases essential skills in data wrangling, visualization, and strategic thinking.

The data used in this project is publicly available data from the FitBit Fitness Tracker dataset on Kaggle linked here

Step 1. Ask

(Back to table of contents)

1.1 Ask: Identifying stakeholders and their expectations

In this step, I have to define the problem I are trying to solve; however, the first step would be to identify the stakeholders and make sure their expectations are understood. Based on provided information in the case study, the project’s stakeholders and their expectations are identified

Stakeholders and their expectations

Stakeholder	Stakeholder Type	Stakeholder Group	Position / Organizational Unit	Expectations
Urška Sršen	Primary stakeholder	Executive team	Cofounder and Chief Creative Officer	- Analyze smart device usage data for insights on current usage - Explore new growth opportunities - Propose high-level recommendations for marketing strategy
Sando Mur	Primary stakeholder	Executive team	Cofounder and Mathematician	- Analyze smart device usage data for insights on current usage - Explore new growth opportunities - Propose high-level recommendations for marketing strategy
Bellabeat marketing analytics team	Secondary stakeholder	Data analysis team	Organizational unit	- Focus on the business task - Collaborate effectively to find new data angles - Support the executive team with actionable insights

Focusing on stakeholder expectations enables data analysis team to understand project goals, improve communication, and build trust

1.2 Ask: Statement of business task

(Back to table of contents)

After identifying stakeholder expectations, it is time to state the business task clearly. Before starting the project, it is necessary to ask a couple of questions, create a summary of the key information and communicate it with the stakeholders

Bellabeat case study information summary

• What is the problem? - Bellabeat believes that there are more opportunities for growth. The marketing analytics team should investigate other non-Bellabeat smart devices, identify potential trends and consumer behaviors, and influence Bellabeat marketing strategy using the gained insight
• Can it be solved with data? - If so, what data? Yes. ‘FitBit Fitness Tracker Data’; it is a public domain data made by Mobius that explores smart device users’ daily habits
• Where is this data? - Does it exist, or do you need to collect it? Are you using private data that someone will need to give you access to, or publicly available data? It exists. It is a public domain data, and the executives encouraged marketing analytics team to use it
• Who are the relevant sponsors and stakeholders for this project? - Who is involved, and how? The executive team is the project sponsors; primary and secondary stakeholders are also identified in previous slide
• What are the boundaries for your project? - The analysis is limited to suggested dataset and proposing high-level recommendations about Bellabeat marketing strategy using gained insights

Bellabeat statement of business task

Analyze non-Bellabeat smart devices data (in this case, FitBit), and identify customer behavior and potential trends (insights)

Apply the insights to Bellabeat products and propose high-level recommendations to guide the company’s marketing strategy

At the end of the ‘Ask’ phase:

• ✔ Project key stakeholders and their expectations are identified

• ✔ Business tasks are set, and the data analysis team knows exactly what it is being asked to accomplish

Step 2. Prepare:

(Back to table of contents)

2.1 Prepare: Dataset introduction

The ‘Prepare’ phase ensures that you have all of the data you need for your analysis and that you have credible and useful data; an introduction to the dataset might be a good starting point.

FitBit’ dataset introduction

• The project founders encouraged to use ‘FitBit fitness tracker data’ made available by Mobius on Kaggle to address the business tasks
• The dataset contains personal fitness data from thirty eligible FitBit users consented to the submission of personal tracker data, including physical activity, steps, heart rate, and sleep monitoring that can be used to explore users’ habits
• According to Mobius on Kaggle, the dataset generated by respondents to a distributed survey via Amazon Mechanical Turk between 2016-12- 03 to 2016-12-05
• Columns information available on Fitbit's data dictionary.
• ‘FitBit’ dataset has CC0 license which allows creators to give up their copyright and put their works into the worldwide public domain. CC0 allows users to distribute, remix, adapt, and build upon the material in any medium or format, with no conditions
• The database contains 18 files in .csv format

2.2 Prepare: Does the data ROCCC?

(Back to table of contents)

Fitbit’ dataset ROCCC analysis

Perspective	Description	Notes	Status
Reliable	Accurate, complete, and unbiased data	The data seems accurate as it is collected from fitness tracker smart devices. However, it contains only 30 respondents, so it is not complete and could be biased.	Low
Original	Validating second or third-party data with original source	The dataset is generated by respondents to a distributed survey via Amazon Mechanical Turk, which is third-party data (even though reliable).	Medium
Comprehensive	All critical information needed to answer the questions is available	An acceptable range of wellness parameters, including physical activity intensity, steps taken, heart rate, sleep monitoring, calories used, weight record, etc., are recorded; however, in a small sample size.	Medium
Current	Up-to-date and relevant data	Provided data dates back to 2016. Therefore, it cannot be considered up-to-date, and the insights gained could be irrelevant now.	Low
Cited	Who created the dataset? Is this dataset from a credible organization?	Data is collected using a survey via Amazon Mechanical Turk. Lack of further information somewhat blurs the credibility of the dataset.	Medium

What is the ROCCC approach? A good dataset should be Reliable, Original, Comprehensive, Current, and Cited. Actually, it's ROCCC

2.3 Prepare: Dataset limitations & data selection

(Back to table of contents)

The two most important limitations of ‘FitBit’ dataset are the dataset small sample size (30 respondents), and the outdated data (dataset dates back to 2016)

FitBit’ dataset limitations

• The smallest sample size for which the Central Limit Theorem (CTL) is still valid would be 30.
• The sample size for some ‘FitBit’ datasets (e.g., sleep monitoring and weight record) is below 30 ‘FitBit’ dataset was generated using a survey in 2016 (9 years ago). Considering the potential changes in consumer habits, the insights gained through analysis could be misleading now (outdated data)
• There is missing data in some datasets (e.g., daily activity and sleep monitoring)
• There is no evidence to verify the validity of the survey, and consequently the integrity of the collected data

Considering the data collection approach (survey) and the fact that I are unable to communicate with stakeholders, modifying the dataset would be impossible at the moment

Project data selection

From the 18 provided datasets, 10 were selected 8 datasets were excluded, mainly because they had duplicate variables across different datasets or they did not match the project context (too many details, where high-level insights are required)

• ❌ Excluded Datasets:

  • dailyCalories_merged (duplicate variables)
  • dailyIntensities_merged (duplicate variables)
  • dailySteps_merged (duplicate variables)
  • heartrate_seconds_merged (too many details)
  • minuteCaloriesWide_merged (too many details)
  • minuteIntensitiesWide_merged (too many details)
  • minuteSleep_merged (vague data)
  • minuteStepsWide_merged (too many details)

At the end of the ‘Prepare’ phase:

✔Required data is collected and stored
✔The credibility of the data is determined
✔The data limitations are recognized

Step 3. Process:

(Back to table of contents)

3.1 Process: Examining the datasets

Now that I decided to continue the project using the current datasets – despite the limitations – in order to answer the business tasks, I will need to clean the data, so that the analysis will be error-free. The first step would be to import and examine the selected datasets

Importing the project datasets RStudio was used to get an overview of the datasets. The following packages were used during analysis: tidyverse and lubridate

library(tidyverse) # For data manipulation
library(skimr) # For data summary
library(janitor) # For cleaning column names
library(lubridate) # For date manipulation
library(readr) # For reading CSV files
library(dplyr) # For data manipulation

The selected datasets were imported using R. Below, you can find the codes used to import the dataset

dailyActivity     <- read_csv("data/dailyActivity_merged.csv")

hourlyCalories    <- read_csv("data/hourlyCalories_merged.csv")
hourlyIntensities <- read_csv("data/hourlyIntensities_merged.csv")
hourlySteps       <- read_csv("data/hourlySteps_merged.csv")

minuteCalories    <- read_csv("data/minuteCaloriesNarrow_merged.csv")
minuteIntensities <- read_csv("data/minuteIntensitiesNarrow_merged.csv")
minuteMETs        <- read_csv("data/minuteMETsNarrow_merged.csv")
minuteSteps       <- read_csv("data/minuteStepsNarrow_merged.csv")

sleepInfo         <- read_csv("data/sleepDay_merged.csv")
weightInfo        <- read_csv("data/weightLogInfo_merged.csv")

Next, it is necessary to examine the imported datasets and decide how to process them

3.2 Process: Examining the datasets

(Back to table of contents)

A summary of the imported datasets is provided below:

Dataset	Description	Number of variables	Variables	Number of rows	Number of respondents
dailyActivity_merged	Data on users’ daily variables	15	Id, ActivityDate, TotalSteps, TotalDistance, TrackerDistance, LoggedActivitiesDistance, VeryActiveDistance, ModeratelyActiveDistance, LightActiveDistance, SedentaryActiveDistance, VeryActiveMinutes, FairlyActiveMinutes, LightlyActiveMinutes, SedentaryMinutes, Calories	940	33
hourlyCalories_merged	Calories consumed by users in hourly intervals	3	Id, ActivityHour, Calories	22,099	33
hourlyIntensities_merged	Intensity of user’s activity in hourly intervals	4	Id, ActivityHour, TotalIntensity, AverageIntensity	22,099	33
hourlySteps_merged	Number of steps taken by users in hourly intervals	3	Id, ActivityHour, StepTotal	22,099	33
minuteCaloriesNarrow_merged	Calories consumed by users in minute intervals (long format)	3	Id, ActivityMinute, Calories	1,325,580	33
minuteIntensitiesNarrow_merged	Intensity of user’s activity in minute intervals (long format)	3	Id, ActivityMinute, Intensity	1,325,580	33
minuteMETsNarrow_merged	METs data of users in minute intervals (long format)	3	Id, ActivityMinute, METs	1,325,580	33
minuteStepsNarrow_merged	Number of steps taken by users in minute intervals (long format)	3	Id, ActivityMinute, Steps	1,325,580	33
sleepDay_merged	Sleep variables of users in daily intervals	5	Id, SleepDay, TotalSleepRecords, TotalMinutesAsleep, TotalTimeInBed	413	24
weightLogInfo_merged	Weight variables of users in daily intervals	8	Id, Date, WeightKg, WeightPounds, Fat, BMI, IsManualReport, LogId	67	8

3.3. Process: Data cleaning and verification

(Back to table of contents)

Now that I learned about the datasets, it is time to focus on the data cleaning and verification processes. They ensure that the data is clean and structured into a suitable format that makes it easy to analyze; the following steps are taken to make it happen.

Data Cleaning Checklist

Step	Description
Check preliminary requirements	- Review business tasks and objectives - Back up data prior to cleaning - Fix errors at the source when possible
Remove unwanted data	- Remove inaccurate, incorrect, and duplicate data
Check data accuracy and correctness	- Validate min/max values, binary data, numeric values (dates, percentages, etc.) - Check spelling, variable names, extra spaces, capitalization, and punctuation
Manage formatting	- Standardize text font, size, color, bold, italic, etc. (as needed for reporting or analysis)
Manage missing values	- Decide how to handle nulls (e.g., impute, remove, or ignore based on context)
Check for merging/splitting	- Identify opportunities to merge or split columns to resolve inconsistencies and misalignments
Check for misfielded values	- Ensure values are entered in the correct fields (e.g., no country names in a 'City' column)
Check data length	- Verify data length where applicable (e.g., year should be 4 digits)

3.4 The data cleaning and verification details

(Back to table of contents)

Based on the previous checklist, the data cleaning and verification were done step by step.

The following describes the details of each step

Remove unwanted data:

Considering the business task, there was irrelevant data in some datasets; thus, the related columns were removed. Below, you can find the mentioned dataset and the removed variables (columns):

Dataset	Action/Change
weightLogInfo_merged	Removed variable logId; removed variable Fat (only 2 records had values, rest were empty); changed column name IsManualReport to Report Type; changed Report Type values from True/False to Manual/Automatic
dailyActivity	Changed column name ActivityDate to Date; changed column name TotalSteps to Steps; removed columns TrackerDistance, LoggedActivitiesDistance, and SedentaryActiveDistance
sleepInfo	Changed column name SleepDay to Date; subtracted TotalMinutesAsleep from TotalTimeInBed to create new column TimeAwake; removed column TotalSleepRecords

Important: All MET values exported from Fitabase are multiplied by 10. Please divide by 10 to get accurate MET values

Identifying duplicate records and removing them is the next step:

The following R code is used to examine how many duplicate rows are there in the sleepDay_merged dataset, and the code was used similarly in all selected datasets; there were only 3 duplicates in the sleepDay_merged dataset and no other duplicates were found.

# Check for duplicates
sum(duplicated(sleepInfo))

To remove the duplicates, the unique() function was used, and to verify the removal of the duplicates, the above code was used again; this time, there was no duplicate.

# verify the removal of the duplicates
sleepInfo <- unique(sleepInfo)

To identify NA values, the following code was used for each selected dataset(here, sleepDay_merged dataset); however, no NA value was found.

# To identify NA values
sum(is.na(sleepInfo))

The following describes the details of each steps:

Data accuracy and correctness:

• Min and Max of each variable were checked (e.g., active minute, active distance, and weight columns should not have any negative value, and the values should be in a reasonable range)
• Dates and times were also examined to be in the proper format. The data types were “character”; therefore, some modifications would be required (discussed in next slides)
• The data spelling and naming columns (e.g., using a single name for all date (ActivityDate), time (ActivityTime), and the total steps taken (TotalSteps) columns in different datasets), capitalization, and any potential extra space were checked and removed

Data formatting: the datasets were checked to have a consistent format (e.g., font type, font color, font size, etc.)

Missing and misfielded values: there were missing values under some of the columns; in some cases, the variables were removed (above table); in other cases, I had to just ignore them, as I did not have access to the respondents to complete the missing values. No misfielded value was found

Length of data: The Id columns were checked in all datasets to have exactly ten digits and no error was found

Check for merging/splitting possibilities: Merging

• According to the selected datasets characteristics table (slides 9 and 10), 3 datasets (hourlyCalories_merged, hourlyIntensities_merged and hourlySteps_merged) have two common variables (Id, ActivityHour) and an equal number of rows. The 3 datasets merged into a single one

library(dplyr)

hourlyParameters <- left_join(hourlyCalories, hourlyIntensities, by = c("id", "activity_hour"))
hourlyParameters <- left_join(hourlyParameters, hourlySteps, by = c("id", "activity_hour"))

• Also, 3 datasets (dailyActivity_merged, sleepDay_merged, and weightLogInfo_merged) contain costumers daily data. Therefore, it would be possible to merge them into a single one ( dailyParameters_merged ) using R merge() function.

dailyParameters <- merge(dailyActivity, sleepInfo, by = c("id", "date"), all.x = TRUE, all.y = FALSE)
dailyParameters <- merge(dailyParameters, weightInfo, by = c("id", "date"), all.x = TRUE, all.y = FALSE)

• Similarly, 4 datasets (minuteCaloriesNarrow_merged, minuteIntensitiesNarrow_merged, minuteMETsNarrow_merged, and minuteStepsNarrow_merged) have two common variables (Id, ActivityMinute) and an equal number of rows. Therefore, to merge the 4 datasets into a single one (minuteParameters_merged), the following R codes were used:

minuteParameters <- left_join(minuteCalories, minuteIntensities, by = c("id", "activity_minute"))
minuteParameters <- left_join(minuteParameters, minuteMETs, by = c("id", "activity_minute"))
minuteParameters <- left_join(minuteParameters, minuteSteps, by = c("id", "activity_minute"))

• By merging the above datasets, the total number of selected datasets decreased from 10 to 3 following datasets:
  • dailyParameters_merged.csv
  • hourlyParameters_merged.csv
  • minuteParameters_merged.csv

3.5 Data merging/splitting possibilities

(Back to table of contents)

Check for merging/splitting possibilities: Splitting

• All datasets had a column with date and/or time data. These columns were split into two date and time columns . For example, the following R codes were used for the minuteParameters_merged.csv dataset (which was created in the merging process – the previous slide) to split the activity_minute column (containing date and time data) into two Date and Time columns. An extra column (Weekday) is also inserted to convert the dates to weekdays:

minuteParameters <- minuteParameters %>%
  mutate(activity_minute = parse_date_time(activity_minute, "%m/%d/%Y %I:%M:%S %p"),
         Date = as.Date(activity_minute),
         Time = format(activity_minute, "%H:%M:%S"),
         Weekday = weekdays(Date))

In the sleepDay_merged dataset, the time for all records was 12:00:00 AM. Therefore, before merging it with the dailyActivity_merged dataset (which was discussed in previous slide), the date-time format was changed to the date only using the following R codes:

sleepInfo <- sleepInfo %>%
 rename(Date=SleepDay) %>% 
 mutate(Date=as.Date(Date,format="%m/%d/%Y")) sleepInfo$Weekday <- weekdays(sleepInfo$Date)

Step 4. Analyze:

(Back to table of contents)

4.1 Analyze: General statistics

Putting data to work in the “Analyze” step. It would be necessary to perform calculations, identify trends and relationships, uncover new insights and discover potential solutions to the business tasks.

Initial analysis of the datasets: Codes

First, general statistics were extracted, including the average, standard deviation, minimum, maximum, and percentiles of some of the selected variables in each dataset

I started with dailyParameters_merged dataset. similar process was repeated for the hourlyParameters_merged_splitted and the minuteParameters_merged_splitted datasets

dailyParameters %>% 
select(steps,total_distance, very_active_minutes, fairly_active_minutes, lightly_active_minutes, sedentary_minutes, calories, total_minutes_asleep, total_time_in_bed, weight_kg, bmi) %>% 
summary

Initial analysis of the datasets: Daily parameters

The initial analysis of the selected variables of the dailyParameters_merged dataset are as follows

The analysis highlights:

The daily average of the total steps is 7,638; it is less than 10,000 steps that most fitness tracking devices recommend I take daily. However, according to US Department of Health and Human Services guidelines, a range between 7,000 to 13,000 steps a day will help to get the full benefits that exercise, including protecting against diseases like cancer and diabetes and helping with weight loss

Users travel approximately an average distance of 5.5 kilometers (5.49 kms) a day

The maximum daily steps taken is 36,019 which is nearly equal to 28 kilometers daily total distance

The population spend nearly 16.5 hours in sedentary time, which is surprisingly high; according to British Heart Foundation, adults of working age in England average about 9.5 hours per day of sedentary time. Users also had an average of 3.2 hours of light activity, and only half an hour of active time (fairly and very active time)

The average calories burned (2,304) seems reasonable. According to the Dietary Guidelines for Americans 2020–2025, the average adult woman expends roughly 1,600 to 2,400 calories per day, while the average adult man expends 2,000 to 3,000

Data shows that while people spend an average of more than 7.5 hours in bed, they sleep about 7 hours a day. That means they are awake in bed for half an hour on average. It also seems normal, as sleep foundation suggests 7-9 hours of daily sleep for adults with 26-64 years old

The average BMI (a person’s weight divided by the square of height) is 25.19. According to Center for Disease Control and prevention, BMI between 25.0 to 29.9 falls within the overweight range

Initial analysis of the datasets: Hourly and minute parameters

The initial analysis of the selected variables of the hourlyParameters_merged_splitted dataset are as follows(RStudio output)

The data for Total intensity and Average intensity is not clear, mainly because the unit for the mentioned variables is not stated. However, these variables will be used in our next analysis

Users took about 320 steps and burned an average of 97 calories per hour, which seems to be in line with the average daily total steps (7,638) and average daily calories burned (2,304)

The initial analysis of the selected variables of the minuteParameters_merged_splitted dataset are as follows (RStudio output)

The metabolic equivalent of task (MET) is the ratio of the work metabolic rate to the resting metabolic rate. For example, one MET is the rate of energy expenditure while at rest. A four MET activity expends four times the energy used by the body at rest. If a person does a 4 MET activity for 30 minutes, he or she has done 4 x 30 = 120 MET-minutes (or 2.0 MET-hours) of physical activity

The data shows that the average MET-minutes is 1.469. Considering the definition of MET, this could correspond to a light activity with a MET value of 2 (such as walking slowly at ~3 km/h) performed for about 0.73 minutes (since 2 × 0.73 ≈ 1.469), or other combinations of very brief light activities. This aligns with generally low activity levels in the dataset.

The average total steps taken by users per hour is close to 5.3 steps, and the number of calories burned is 1.6

4.2 Analyze: Identifying trends and relationships

1. Activity time vs. Calories

(Back to table of contents)

Relationships between the Activity level (very active, fairly active, lightly active, and sedentary), and the Activity duration was examined. For Sedentary activity level, the negative slope of the regression line shows that as I spend more sedentary time, the number of calories consumed decreases

Sedentary time (minutes) vs. number of calories consumed

dailyParameters %>%
  ggplot() +
  geom_point(aes(x = sedentary_minutes, y = calories), color = "darksalmon") +
  geom_smooth(aes(x = sedentary_minutes, y = calories), method = lm, se = FALSE, color = "cornsilk4") +
  labs(
    title = "Calorie vs. Sedentary Time",
    subtitle = "The relationship between the sedentary time of the users and the number of calories burned",
    caption = "Data collected from 33 users",
    x = "X: Sedentary time (minutes)",
    y = "Y: Number of calories burned"
  ) +
  annotate("text", x = 50, y = 2750, label = "Regression line", color = "cornsilk4", size = 3, fontface = "bold") +
  theme_minimal()

The p-value is equal to 0.001; a typical threshold is 0.05, and anything smaller is considered statistically significant

2.Activity time vs. Calories

(Back to table of contents)

Very, fairly, and lightly active time (minutes) vs. number of calories consumed Sample codes for “Very active” level

#veryactiveminutes
dailyParameters %>%
  ggplot() +
  geom_point(aes(x = very_active_minutes, y = calories), color = "darkolivegreen3") +
  geom_smooth(aes(x = very_active_minutes, y = calories), method = lm, se = FALSE, color = "cornsilk4", size = 0.5) +
  labs(
    title = "Calorie vs. very_active_minutes",
    subtitle = "Relationship btw the very_active_minutes of the users and the number of calories burned",
    caption = "Data collected from 33 users",
    x = "X: very_active_minutes (minutes)",
    y = "Y: Number of calories burned"
  ) +
  annotate("text", x = 50, y = 2750, label = "Regression line", color = "cornsilk4", size = 3, fontface = "bold") +
  theme_minimal()

If you are more active (“very” or “fairly”), you may burn more calories for less activity time

As mentioned earlier, the average adult woman expends roughly 1,600 to 2,400 calories and the average adult man expends 2,000 to 3,000. The diagrams confirm that the majority of data points are concentrated between the above range (1600-3000 calories/day)

3.Sedentary time vs. Sleep

(Back to table of contents)

Data shows that sedentary time has an inverse effect on sleep time. Therefore, it seems that more physical activity can improve users' sleep time

dailyParameters %>%
  drop_na() %>%
  ggplot(aes(x = sedentary_minutes, y = total_minutes_asleep)) +
  geom_point(color = "darksalmon") +
  geom_smooth(method = "lm", se = FALSE, color = "cornsilk4",size=0.5) +
  labs(
    title = "Sedentary time vs. Total sleep time",
    subtitle = "The relationship between users' sedentary time and their total sleep time",
    caption = "Data collected from 33 users",
    x = "X: Users' sedentary time (minutes)",
    y = "Y: Users' total sleep time (minutes)"
  ) +
  theme_minimal()

The diagram shows that as users’ sedentary time increases, total sleep time decreases (negative slope)

In other word, lack of physical activity seems to have a negative effect on users’ sleep time

4. Activity time vs. BMI

(Back to table of contents)
Activity duration (minutes) of the users at different levels vs. BMI

A linear regression model describes the relationship between a dependent variable (y), and one or more independent variables (x).

To examine the relationship between users' different activity levels and their weight, BMI was used as a better indicator—mainly because it considers the users' height.

Here, p-value and R-squared (R²) were considered in each activity level to determine any potential relationship.

P-value is used in hypothesis testing to help decide whether to reject the null hypothesis (primarily, the one that indicates there is no relationship between the variables). The smaller the p-value (a typical threshold is 0.05; anything smaller is considered statistically significant), the more likely to reject the null hypothesis.

R-squared is a goodness-of-fit measure for linear regression, and always is between 0–1. Usually, the larger the R², the better the regression model fits your observations.

Sample codes for “Very active” level

#veryActiveBMI_correlation
veryActive_BMI_correlation <- lm(very_active_minutes ~ bmi, data = dailyParameters)

summary(veryActive_BMI_correlation)

The relationship can be considered statistically significant at the ‘Fairly active’ and “Lightly active” levels; at other activity levels, no correlation is seen

5.Share of daily activity levels

(Back to table of contents)
While users spend most of their time in sedentary state, the duration of moderate-to-vigorous physical activity is largely meets World Health Organization guidelines

Description	Moderate-intensity activity	Vigorous-intensity activity
WHO minutes/week	150-300	75-100
WHO minutes/day	21-42	11-21
Users compliance rate (%)	82% (27 out of 33 users)	88% (29 out of 33 users)

Over 80% of users had enough daily moderate, and vigorous-intensity physical activity

The users spent the majority of their time in sedentary mode (81% of the time), while they were active (very or fairly) only 3% of the time

6.Total steps vs. Calories

(Back to table of contents)
There is a positive relationship between the total steps taken and the number of calories burned (p-value < 0.05). In other words, as the number of steps taken increases, the calorie consumption increases.

Total steps taken by the users vs. number of calories consumed

The bar chart shows that there is a positive significant correlation between the user's total steps and the number of calories burned (regression line). The histogram on the right shows the distribution of total daily steps taken. The more daily steps you take, the more calories you burn

7.Activity time vs. Total steps

(Back to table of contents)
As the plots suggest, the longer sedentary time, the fewer steps users take; all other relationships are positive. In addition, it is observed that as the activity level increases, less time is required to take a certain number of steps.

Activity duration (minutes) at different levels vs. total number of steps taken by the users

8.Analyze: Identifying trends and relationships – Activity time by the weekdays

(Back to table of contents)
Monday has the highest sedentary time, and Thursday has the least. On the other hand, Saturday has the highest active time, and Sunday has the least

The average sedentary time and the average active time (the sum of `Very active’, ‘Fairly active, and ‘Lightly active’ time averages) by the weekdays were calculated and plotted using Microsoft Excel pivot table and charts

The above charts show that sedentary time is among the lowest on Saturday (left chart), while active time is at the peak (right chart)

9.Number of steps and degree of intensity per hour

(Back to table of contents)
Understanding user behavior at different hours of the day could be useful. Therefore, the number of steps taken and the intensity of their activity per hour have been analyzed and plotted

Total number of steps and total intensity by hours

#Total number of steps and total intensity by hours
hourlyParameters <- hourlyParameters %>%
  mutate(hour = hour(mdy_hms(activity_hour)))

# Then group by hour instead of raw timestamp
hourlyStepsSummary <- hourlyParameters %>%
  group_by(hour) %>%
  summarise(avgTotalSteps = mean(step_total, na.rm = TRUE))

hourlyIntensitySummary <- hourlyParameters %>%
  group_by(hour) %>%
  summarise(avgTotalIntensity = mean(total_intensity, na.rm = TRUE))

# Scale factor for secondary axis
scale_factor <- 0.05 * max(hourlyStepsSummary$avgTotalSteps, na.rm = TRUE)

# Plot
ggplot() +
  geom_col(data = hourlyStepsSummary, aes(x = factor(hour), y = avgTotalSteps), fill = "darksalmon") +
  geom_line(data = hourlyIntensitySummary,
            aes(x = factor(hour), y = avgTotalIntensity * scale_factor, group = 1),
            color = "dimgrey") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 90, face = "bold")) +
  labs(
    title = "Total Steps and Intensity per Hour",
    subtitle = "The total number of steps taken and the degree of intensity per hour",
    caption = "Data collected from 33 users",
    x = "Hour of Day",
    y = "Total Steps (bars) / Degree of Intensity (line)"
  ) +
  scale_y_continuous(
    sec.axis = sec_axis(~ . / scale_factor, name = "Intensity")
  )

As it may seem, the total steps taken and the degree of intensity follow a relatively similar pattern. Also, data shows that most steps are taken by users in the afternoon and evening

TotalSteps~TotalIntensity: p-value < 0.05, R2: 0.8028

The histogram shows that as the day begins, the number of steps taken by users increases gradually until 5 PM, and then **decreases until the end of the day **(bell-shaped distribution); interestingly, the degree of intensity follows the same pattern

Two significant intervals are 12-2 PM and 5-7 PM (probably after working hours), where the number of steps (and of course, the degree of intensity) is higher than the other hours of the day

Relationship between total steps taken and sedentary minutes

Inverse Relationship:

More active users (those with higher step counts) are less sedentary — a logically expected pattern.

This supports the idea that increasing daily steps can help reduce sedentary behavior.

Two Activity Clusters:

There appear to be two main bands of sedentary behavior (one higher, one lower) even for the same range of steps, suggesting:

Some users may be getting in steps via short bursts (e.g., workouts) but still sitting for long durations.

Others may be spreading out their activity better through the day.

Outliers:

A few users have very high steps (20k–30k+) and still high sedentary minutes, which may reflect unusual routines (e.g., workers with a long commute but active jobs).

Relationship between time in bed and minutes asleep

observations:

Strong linear correlation:

Most data points lie close to the line, indicating that as minutes asleep increase, time in bed increases almost proportionally.

This suggests a consistent pattern of sleep behavior across most users.

Efficiency observed: The slope near the main cluster implies high sleep efficiency (users are asleep for most of their time in bed).

The closeness of many points to the diagonal suggests users spend only a small portion of time in bed awake.

Step 5 Share

(Back to table of contents)

During the “share” phase, it is necessary to visualize the findings and determine the best way to communicate them to stakeholders.

Step 6 Act

(Back to table of contents)

6.1 Act: Insights

After data analysis, creating meaningful visualizations, and sharing them with the stakeholders, I have to act on the findings. The deliverables should be prepared, and the business tasks should be answered, including high-level recommendations based on insights gained

• Number of steps

  • The daily average of total steps taken by users is 7638. That is less than the 10,000 steps most references recommend
  • Activity duration and total steps taken by users have a negative relationship at the sedentary level. A positive correlation can be seen in the other three activity levels ("very active", “fairly active" and “lightly active")
  • The total steps taken and the degree of intensity are highly correlated (p-value < 0.05, R2: 0.802)
  • Users take more steps in the afternoons (12-2 PM) and evenings (5-7 PM) than at other times of the day

• Calories

  • The average calories burned (2,304) seems reasonable, compared to the official guidelines (1,600 to 3,000). Most data points are also concentrated between the above range
  • There is a positive significant correlation between the activity duration and the number of calories burned in “very active”, “fairly active”, and “lightly active” levels. This means that if you are more active, you can probably burn more calories for less activity duration
  • At the “sedentary” level, the relationship is negative
  • The calorie consumption increases as the number of steps taken increases

• Activity levels

  • Users spent 81% of their time in sedentary mode (lack of physical activity), which is surprisingly high. They were active (very or fairly) only 3% of the time
  • Sedentary time seems to have a negative effect on sleep time. Therefore, more physical activity can probably improve users' sleep quality and time
  • Users spend a daily average of more than 7.5 hours in bed (7 hours of sleep and half an hour in bed)
  • Monday has the highest sedentary time, and Thursday has the least. Saturday has the highest active time, and Sunday has the least

• Weight

  • The average BMI is 25.19 which falls into the "overweight” category" (25.0 to 29.9)
  • The average of metabolic equivalent of task (MET) is 14.69, which represents light activity (e.g., walking at a slow pace) for a short duration (less than 10 minutes)
  • It appears to be little or no correlation between the user’s “activity time” and “weight”; only at the ‘Fairly active’ and “Lightly active” levels, the correlation seems to be tatistically significant (p-value < 0.05)

6.2 Act: Recommendations

(Back to table of contents)

Based on insights gained from data analysis, it would be possible to apply the insights to Bellabeat products and propose high-level recommendations to guide the company’s marketing strategy (the business task)

6.2.1 Recommendations

• The following steps are a general process, the Bellabeat is recommended to follow for each insight discussed earlierin 6.1:
  • Share each insight with current and potential customers using smart marketing methods (digital marketing, marketing campaigns, classic ads, etc.)
  • Highlight the potential risks, including exposure to various diseases (cancer, diabetes, etc.), weight gain, etc., that an unhealthy lifestyle could cause, preferably with statistics
  • Improve the Bellabeat app and include a *user-friendly notification system to provide users with customized alerts. The objective is to inform users about their health habits, and encourage them to improve them
  • Promote the benefits of using Bellabeat products and the app to avoid negative health habits

6.2.2 Recommendations

• Bellabeat can focus on the “activity level”, “activity duration”, and “total steps” of the users, and improve related products and functions of the app. It should be highlighted that users spent more than 80% of their time in sedentary mode, while they were active (very or fairly) only 3% of the time. They took less than the 10,000 steps (7638 steps) that most guidelines recommend, and therefore, had less than average activity intensity (high positive correlation between “total steps” and “intensity”)
• The company can organize social challenges and special events, targeting days when sedentary time may be at its peak (Mondays), active time seems to be at its lowest (Sundays), or hours of a day when the number of steps taken is less, using the Bellabeat app. It can also offer incentives such as some company products, special discounts, free membership, etc. to the winners
• Bellabeat should explain the importance of physical activity on sleep quality and sleep duration. Bellabeat needs to improve sleep tracking function in both its products (e.g., required hardware) and the app. It should emphasize the fact that the usage of Bellabeat’s smart devices could help customers increase their activity levels and monitor their sleep habits
• The average calorie intake of users appears to be in the acceptable range (2,304 calories per day, which is within the recommended range of 1,600 to 3,000). However, the average BMI (25.19) is in the "overweight" category, and the average MET (14.69) indicates very light activity for short periods of time. Therefore, it may be a good idea to encourage users to actively manage their weight/BMI (weight management practices)
• The direct positive correlation between “activity duration” and “calorie consumption” means that if you are more active, you can burn more calories for less activity duration. Bellabeat smart devices and the app can help customers monitor their activity level and try to improve it
• Further analysis needs to be done to get more accurate and reliable results.It is mainly because there were some data limitations (i.e. the small sample size for some ‘FitBit’ datasets, outdated datasets and potential changes in consumer habits, missing or manual data input in some cases, etc.), and I were unable to modify it properly due to the data collection approach (survey in 2016)

Skills Used

(Back to table of contents)

Data Analysis Skills in R

• Data Cleaning: Preparing raw data for analysis by handling missing values, outliers, and inconsistencies.

• Data Visualization:Creating informative and aesthetically pleasing visualizations using ggplot2.

• Statistical Analysis:Performing statistical tests and modeling to draw insights from data.

• ggplot2:Utilizing the ggplot2 package for advanced data visualization.

• dplyr:Using dplyr for efficient data manipulation and transformation.

• Tidying Data:Structuring data in a tidy format for easier analysis and visualization.

• R Markdown:Documenting and presenting data analysis using R Markdown for reproducible research.

• Exploratory Data Analysis (EDA):Analyzing data to summarize its main characteristics, often with visual methods.

• Data Importing and Exporting:Reading data from various sources and writing results to files.

• Date Functions: Handling and manipulating date and time data.

Limitations and future scope

(Back to table of contents)

To begin with, it must be highlighted that the public domain dataset i.e., the meta- data used for this case study has its limitations.

There is no demographic information on the users, there is no clarity of the Fitbit devices the users wore and if at all they were same devices, there is a variance in the metadata as a result.

Also, the way the data was originally captured is not qualitative enough to have analysed and there- fore required serious cleaning and manipulation prior to the actual analysis.

For the purpose of future work, the metadata can be combined with other relevant available data to uncover deeper insights into trends, and usage patterns of the users which definitely will lead to better recommendation for Bellabeat’s product portfolio as well as their marketing team’s strategy

References

(Back to table of contents)

1. Wickham et al., (2019). Welcome to the tidyverse. Journal of Open Source Soft- ware, 4(43), 1686, https://doi.org/10.21105/joss.01686.
2. Garrett Grolemund, Hadley Wickham (2011). Dates and Times Made Easy with lubridate. Journal of Statistical Software, 40(3), 1-25, https://www.jstatsoft.org/v40/i03/.
3. Kirill Müller (2021). hms: Pretty Time of Day. R package version 1.1.1, https://CRAN.R-project.org/package=hms.
4. Hadley Wickham and Dana Seidel (2020). scales: Scale Functions for Visualiza- tion. R package version 1.1.1, https://CRAN.R-project.org/package=scales.
5. Sam Firke (2021). janitor: Simple Tools for Examining and Cleaning Dirty Data. R package version 2.1.0, https://CRAN.R-project.org/package=janitor.
6. Elin Waring, Michael Quinn, Amelia McNamara, Eduardo Arino de la Rubia, Hao Zhu and Shannon Ellis (2021). skimr: Compact and Flexible Summaries of Data. R package version 2.1.3, https://CRAN.R-project.org/package=skimr.
7. Joshua Kunst (2022). highcharter: A Wrapper for the ‘Highcharts’ Library. R package version 0.9.4, https://CRAN.R-project.org/package=highcharter.
8. Brinton, Julia E, et al. “Establishing Linkages between Distributed Survey Re- sponses and Consumer Wearable Device Datasets: A Pilot Protocol.” JMIR Re- search Protocols, vol. 6, no. 4, 2017, https://doi.org/10.2196/resprot.6513.
9. CDC. "Assessing Your Weight" CDC, September 17, 2020, https://www.cdc.gov/healthyweight/assessing/index.html.
10. CDC. "How much physical activity do adults need?" CDC, October 7, 2020, https://www.cdc.gov/physicalactivity/basics/adults/index.htm.
11. CDC. "How Much Sleep Do I Need?" CDC, March 2, 2017, https://www.cdc.gov/sleep/about/?CDC_AAref_Val=https://www.cdc.gov/sleep/about_sleep/how_much_sleep.html
12. Fitbit. "What should I know about Fitbit sleep stages?" Fitbit, 2022, https://help.fitbit.com/articles/enUS/Helparticle/2163.htm.
13. Möbius. “Fitbit Fitness Tracker Data.” Kaggle, 16 Dec. 2020, https://www.kaggle.com/arashnic/fitbit.
14. Asimakopoulos, Stavros, Grigorios Asimakopoulos, and Frank Spillers. "Moti- vation and user engagement in fitness tracking: Heuristics for mobile healthcare wearables." In Informatics, vol. 4, no. 1, p. 5. MDPI, 2017, https://www.mdpi.com/2227-9709/4/1/5.