Case Study: How Can a Wellness Technology Company Play It Smart?¶

Part of the Google Analytics Professional Certificate¶

by Whoong Zi Wei¶


Bellabeat is a high-tech company founded by Urška Sršen and Sando Mur that manufactures health-focused smart products for women. In this case study, I am presented a scenario where I will be working for Urška Sršen and I have been asked to focus on one of Bellabeat's products and analyze smart device data to gain insight into how consumers are using their smart devices.

In order to help guide Bellabeat in unlocking new growth opportunities, I am requested by Urška Sršen to analyze available consumer data, which for this scenario I am to use a dataset by Fitbit Fitness Tracker Data (Public Domain, dataset made available through Mobius). This dataset contains personal fitness tracker from thirty fitbit users. These thirty Fitbit users had consented to the submission of personal tracker data, which includes minute-level output for physical activity, heart rate, and sleep monitoring. It also includes information about daily activity, steps, and heart rate that can be used to explore users' habits.


Key Stakeholders:¶

  • Urska Srsen - Bellabeat cofounder and CCO
  • Sando Mur - Bellabeat cofounder
  • Bellabeat marketing analytics team

Bellabeat's Key Products:¶

  • Bellabeat app: The Bellabeat app provides users with health data related to their activity. The Bellabeat app connects to their line of smart wellness products.
  • Leaf: Worn as either a bracelet, necklace, or clip; the Leaf tracker connects to the Bellabeat app to track activity, sleep, and stress.
  • Time: A wellness watch that uses smart technology to track the user's activity, sleep, and stress.
  • Spring: A water bottle that tracks daily water intake. The bottle connects to the Bellabeat app to track the users hydration levels.
  • Bellabeat membership: A subscription-based membership program for users. Membership gives users 24/7 access to fully personalized guidance on nutrition, activity, sleep, health and beauty, and mindfulness based on their lifestyle and goals.


Business Objective¶

  1. What are some trends in smart device usage?
  2. How could these trends apply to Bellabeat customers?
  3. How could these trends help influence Bellabeat marketing strategy?

As this dataset has more than 10 csv files available, I am encouraged by Sršen to use as much as I can in order aid my research.

Loading Packages¶

In [1]:
import numpy as np 
import pandas as pd 
import matplotlib.pyplot as plt
import missingno as msno
import seaborn as sns

Prepare Phase¶

The datasets are gathered internally by FitBit, therefore the data can be safely assumed to be unbias and credible. I have also taken a look at the datasets in a spreadsheet software and there doesn't seem to be any problems other than the date formats are not properly converted. I will convert them to datetime formats when the need arises.

There are a total of 18 datasets available. However, I will only need a few of them for this case study. I will also merge some datasets if neccesary as I progress through this case study.¶

In [2]:
pd.options.display.width = None
pd.options.display.max_columns = None

dailyActivity = pd.read_csv('Fitabase Data 4.12.16-5.12.16/dailyActivity_merged.csv')
dailyCalories = pd.read_csv('Fitabase Data 4.12.16-5.12.16/dailyCalories_merged.csv')

minuteSleep = pd.read_csv('Fitabase Data 4.12.16-5.12.16/minuteSleep_merged.csv')
dailySteps = pd.read_csv('Fitabase Data 4.12.16-5.12.16/dailySteps_merged.csv')
heartrateSeconds = pd.read_csv('Fitabase Data 4.12.16-5.12.16/heartrate_seconds_merged.csv')
hourlyCalories = pd.read_csv('Fitabase Data 4.12.16-5.12.16/hourlyCalories_merged.csv')

hourlyIntensities = pd.read_csv('Fitabase Data 4.12.16-5.12.16/hourlyIntensities_merged.csv')
hourlySteps = pd.read_csv('Fitabase Data 4.12.16-5.12.16/hourlySteps_merged.csv')
sleepDay = pd.read_csv('Fitabase Data 4.12.16-5.12.16/sleepDay_merged.csv')
stepsDf = pd.read_csv('Fitabase Data 4.12.16-5.12.16/hourlySteps_merged.csv')
weightLog = pd.read_csv('Fitabase Data 4.12.16-5.12.16/weightLogInfo_merged.csv')
In [3]:
hourlySteps.head()
Out[3]:
Id ActivityHour StepTotal
0 1503960366 4/12/2016 12:00:00 AM 373
1 1503960366 4/12/2016 1:00:00 AM 160
2 1503960366 4/12/2016 2:00:00 AM 151
3 1503960366 4/12/2016 3:00:00 AM 0
4 1503960366 4/12/2016 4:00:00 AM 0
In [4]:
hourlyIntensities.head()
Out[4]:
Id ActivityHour TotalIntensity AverageIntensity
0 1503960366 4/12/2016 12:00:00 AM 20 0.333333
1 1503960366 4/12/2016 1:00:00 AM 8 0.133333
2 1503960366 4/12/2016 2:00:00 AM 7 0.116667
3 1503960366 4/12/2016 3:00:00 AM 0 0.000000
4 1503960366 4/12/2016 4:00:00 AM 0 0.000000
In [5]:
heartrateSeconds.head()
Out[5]:
Id Time Value
0 2022484408 4/12/2016 7:21:00 AM 97
1 2022484408 4/12/2016 7:21:05 AM 102
2 2022484408 4/12/2016 7:21:10 AM 105
3 2022484408 4/12/2016 7:21:20 AM 103
4 2022484408 4/12/2016 7:21:25 AM 101
In [6]:
dailySteps.head()
Out[6]:
Id ActivityDay StepTotal
0 1503960366 4/12/2016 13162
1 1503960366 4/13/2016 10735
2 1503960366 4/14/2016 10460
3 1503960366 4/15/2016 9762
4 1503960366 4/16/2016 12669
In [7]:
dailyCalories.head()
Out[7]:
Id ActivityDay Calories
0 1503960366 4/12/2016 1985
1 1503960366 4/13/2016 1797
2 1503960366 4/14/2016 1776
3 1503960366 4/15/2016 1745
4 1503960366 4/16/2016 1863
In [8]:
# sleepDay.shape (413, 5)
sleepDay.head()
Out[8]:
Id SleepDay TotalSleepRecords TotalMinutesAsleep TotalTimeInBed
0 1503960366 4/12/2016 12:00:00 AM 1 327 346
1 1503960366 4/13/2016 12:00:00 AM 2 384 407
2 1503960366 4/15/2016 12:00:00 AM 1 412 442
3 1503960366 4/16/2016 12:00:00 AM 2 340 367
4 1503960366 4/17/2016 12:00:00 AM 1 700 712
In [9]:
minuteSleep.head()
Out[9]:
Id date value logId
0 1503960366 4/12/2016 2:47:30 AM 3 11380564589
1 1503960366 4/12/2016 2:48:30 AM 2 11380564589
2 1503960366 4/12/2016 2:49:30 AM 1 11380564589
3 1503960366 4/12/2016 2:50:30 AM 1 11380564589
4 1503960366 4/12/2016 2:51:30 AM 1 11380564589
In [10]:
dailyActivity.head()
Out[10]:
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories
0 1503960366 4/12/2016 13162 8.50 8.50 0.0 1.88 0.55 6.06 0.0 25 13 328 728 1985
1 1503960366 4/13/2016 10735 6.97 6.97 0.0 1.57 0.69 4.71 0.0 21 19 217 776 1797
2 1503960366 4/14/2016 10460 6.74 6.74 0.0 2.44 0.40 3.91 0.0 30 11 181 1218 1776
3 1503960366 4/15/2016 9762 6.28 6.28 0.0 2.14 1.26 2.83 0.0 29 34 209 726 1745
4 1503960366 4/16/2016 12669 8.16 8.16 0.0 2.71 0.41 5.04 0.0 36 10 221 773 1863
In [11]:
stepsDf.head()
Out[11]:
Id ActivityHour StepTotal
0 1503960366 4/12/2016 12:00:00 AM 373
1 1503960366 4/12/2016 1:00:00 AM 160
2 1503960366 4/12/2016 2:00:00 AM 151
3 1503960366 4/12/2016 3:00:00 AM 0
4 1503960366 4/12/2016 4:00:00 AM 0
In [12]:
hourlyCalories.head()
Out[12]:
Id ActivityHour Calories
0 1503960366 4/12/2016 12:00:00 AM 81
1 1503960366 4/12/2016 1:00:00 AM 61
2 1503960366 4/12/2016 2:00:00 AM 59
3 1503960366 4/12/2016 3:00:00 AM 47
4 1503960366 4/12/2016 4:00:00 AM 48
In [13]:
weightLog.head()
Out[13]:
Id Date WeightKg WeightPounds Fat BMI IsManualReport LogId
0 1503960366 5/2/2016 11:59:59 PM 52.599998 115.963147 22.0 22.650000 True 1462233599000
1 1503960366 5/3/2016 11:59:59 PM 52.599998 115.963147 NaN 22.650000 True 1462319999000
2 1927972279 4/13/2016 1:08:52 AM 133.500000 294.317120 NaN 47.540001 False 1460509732000
3 2873212765 4/21/2016 11:59:59 PM 56.700001 125.002104 NaN 21.450001 True 1461283199000
4 2873212765 5/12/2016 11:59:59 PM 57.299999 126.324875 NaN 21.690001 True 1463097599000
In [14]:
dailyActivity.head()
Out[14]:
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories
0 1503960366 4/12/2016 13162 8.50 8.50 0.0 1.88 0.55 6.06 0.0 25 13 328 728 1985
1 1503960366 4/13/2016 10735 6.97 6.97 0.0 1.57 0.69 4.71 0.0 21 19 217 776 1797
2 1503960366 4/14/2016 10460 6.74 6.74 0.0 2.44 0.40 3.91 0.0 30 11 181 1218 1776
3 1503960366 4/15/2016 9762 6.28 6.28 0.0 2.14 1.26 2.83 0.0 29 34 209 726 1745
4 1503960366 4/16/2016 12669 8.16 8.16 0.0 2.71 0.41 5.04 0.0 36 10 221 773 1863

After taking a look and understanding the datasets available, I've decided to use the above 11 datasets for my analysis.¶

After having an understanding of these available datasets, I have prepared these questions in order to aid myself in the analysis phase.

  • How active are these users?
  • What factors contribute to the highest calorie burns?
  • Have these users been having a good night sleep? Is there a relationship between how active they are and their sleep schedule?
  • How does the data vary from weekends and weekdays?
In [15]:
dailyActivity.head()
Out[15]:
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories
0 1503960366 4/12/2016 13162 8.50 8.50 0.0 1.88 0.55 6.06 0.0 25 13 328 728 1985
1 1503960366 4/13/2016 10735 6.97 6.97 0.0 1.57 0.69 4.71 0.0 21 19 217 776 1797
2 1503960366 4/14/2016 10460 6.74 6.74 0.0 2.44 0.40 3.91 0.0 30 11 181 1218 1776
3 1503960366 4/15/2016 9762 6.28 6.28 0.0 2.14 1.26 2.83 0.0 29 34 209 726 1745
4 1503960366 4/16/2016 12669 8.16 8.16 0.0 2.71 0.41 5.04 0.0 36 10 221 773 1863

How active are these Users?¶

In [16]:
#convert to datetime
dailyActivity['ActivityDate'] = pd.to_datetime(dailyActivity['ActivityDate'])

# add a new column that indicates the day's name
dailyActivity['day_name'] = dailyActivity['ActivityDate'].dt.day_name()

# add a new column that indicates weekday/weekend
dailyActivity['weekend_weekday'] = np.where(dailyActivity['ActivityDate'].dt.dayofweek > 4, 'Weekend', 'Weekday')

Let's get a brief summary of their daily activities.¶

In [17]:
# steps taken summary
print("Total Steps Summary")
steps = dailyActivity['TotalSteps'].describe()
print(steps)
print('\n')

# byMinutes summary
minutes = dailyActivity[['VeryActiveMinutes', 'FairlyActiveMinutes', 'LightlyActiveMinutes', 'SedentaryMinutes']].describe()
print(minutes)
print('\n')

# Calories and total distance burned summary
calories = dailyActivity[['Calories', 'TotalDistance']].describe()
print(calories)
print('\n')

#Sleep Records, weight, and bmi
print(sleepDay[['TotalSleepRecords', 'TotalMinutesAsleep', 'TotalTimeInBed']].describe())
print('\n')
print(weightLog[['BMI', 'WeightKg']].describe())

Total Steps Summary
count      940.000000
mean      7637.910638
std       5087.150742
min          0.000000
25%       3789.750000
50%       7405.500000
75%      10727.000000
max      36019.000000
Name: TotalSteps, dtype: float64


       VeryActiveMinutes  FairlyActiveMinutes  LightlyActiveMinutes  \
count         940.000000           940.000000            940.000000   
mean           21.164894            13.564894            192.812766   
std            32.844803            19.987404            109.174700   
min             0.000000             0.000000              0.000000   
25%             0.000000             0.000000            127.000000   
50%             4.000000             6.000000            199.000000   
75%            32.000000            19.000000            264.000000   
max           210.000000           143.000000            518.000000   

       SedentaryMinutes  
count        940.000000  
mean         991.210638  
std          301.267437  
min            0.000000  
25%          729.750000  
50%         1057.500000  
75%         1229.500000  
max         1440.000000  


          Calories  TotalDistance
count   940.000000     940.000000
mean   2303.609574       5.489702
std     718.166862       3.924606
min       0.000000       0.000000
25%    1828.500000       2.620000
50%    2134.000000       5.245000
75%    2793.250000       7.712500
max    4900.000000      28.030001


       TotalSleepRecords  TotalMinutesAsleep  TotalTimeInBed
count         413.000000          413.000000      413.000000
mean            1.118644          419.467312      458.639225
std             0.345521          118.344679      127.101607
min             1.000000           58.000000       61.000000
25%             1.000000          361.000000      403.000000
50%             1.000000          433.000000      463.000000
75%             1.000000          490.000000      526.000000
max             3.000000          796.000000      961.000000


             BMI    WeightKg
count  67.000000   67.000000
mean   25.185224   72.035821
std     3.066963   13.923206
min    21.450001   52.599998
25%    23.959999   61.400002
50%    24.389999   62.500000
75%    25.559999   85.049999
max    47.540001  133.500000

Insights from this summary:¶

  • Sedentary average around 991 minutes, or 16 hours.
  • The users on this dataset are mostly lightly active.
  • Users seem to be getting on average roughly 7 hours of sleep, and they mostly sleep one time per day on average.
  • On average, these users take 7637 steps per day, which is lower than what CDC recommends on one of their research. According to a research conducted by the CDC, an adult should ideally take at least 8000 steps per day as it is associated with a 51% lower risk for all-cause mortality.

I recommend Bellabeat to program their smart devices to inform users regarding the ideal total steps to be taken to their users.

Let's see if the day of the week, or weekends or weekdays, have any influence on how users are active.¶

In [18]:
byMinute = dailyActivity.pivot_table(
    values = ['VeryActiveMinutes', 'FairlyActiveMinutes', 'LightlyActiveMinutes', 'SedentaryMinutes'],
    index = 'day_name'
)

byMinute
Out[18]:
FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes VeryActiveMinutes
day_name
Friday 12.111111 204.198413 1000.309524 20.055556
Monday 14.000000 192.058333 1027.941667 23.108333
Saturday 15.201613 207.145161 964.282258 21.919355
Sunday 14.528926 173.975207 990.256198 19.983471
Thursday 11.959184 185.421769 961.993197 19.408163
Tuesday 14.335526 197.342105 1007.361842 22.953947
Wednesday 13.100000 189.853333 989.480000 20.780000
In [19]:
byDistance = dailyActivity.pivot_table(
    values = ['VeryActiveDistance', 'ModeratelyActiveDistance', 'LightActiveDistance', 'SedentaryActiveDistance'],
    index = 'day_name'
)

byDistance
Out[19]:
LightActiveDistance ModeratelyActiveDistance SedentaryActiveDistance VeryActiveDistance
day_name
Friday 3.489127 0.483810 0.001825 1.312937
Monday 3.363083 0.585833 0.002583 1.537333
Saturday 3.617177 0.677339 0.001048 1.514597
Sunday 2.892314 0.618017 0.000661 1.488926
Thursday 3.283129 0.505170 0.002313 1.390476
Tuesday 3.471053 0.593026 0.001447 1.613289
Wednesday 3.256333 0.527067 0.001333 1.633467
In [20]:
byMinute.plot(kind = 'bar', figsize = [17,7])
Out[20]:
<AxesSubplot:xlabel='day_name'>
In [21]:
byDistance.plot(kind = 'bar', figsize = [17,7])
Out[21]:
<AxesSubplot:xlabel='day_name'>

The most and least active based on the type of activity (Fairly, Lightly, etc)¶

In [22]:
print('The Most Active Based on the Type of Activity (Fairly, Lightly, etc)')
print(byMinute.idxmax())
print('\n')
print('The Least Active Based on the Type of Activity (Fairly, Lightly, etc)')
print(byMinute.idxmin())

The Most Active Based on the Type of Activity (Fairly, Lightly, etc)
FairlyActiveMinutes     Saturday
LightlyActiveMinutes    Saturday
SedentaryMinutes          Monday
VeryActiveMinutes         Monday
dtype: object


The Least Active Based on the Type of Activity (Fairly, Lightly, etc)
FairlyActiveMinutes     Thursday
LightlyActiveMinutes      Sunday
SedentaryMinutes        Thursday
VeryActiveMinutes       Thursday
dtype: object

The Most and Least Active Based On Distance¶

In [23]:
print('The Most Active Based On Distance')
print(byDistance.idxmax())
print('\n')
print('The Least Active Based On Distance')
print(byDistance.idxmin())

The Most Active Based On Distance
LightActiveDistance          Saturday
ModeratelyActiveDistance     Saturday
SedentaryActiveDistance        Monday
VeryActiveDistance          Wednesday
dtype: object


The Least Active Based On Distance
LightActiveDistance         Sunday
ModeratelyActiveDistance    Friday
SedentaryActiveDistance     Sunday
VeryActiveDistance          Friday
dtype: object

If compared based on the days of the week, Monday seems to be the least active day on average based on the sedentary being the highest in both in terms of minutes and distance. Based on the data, Saturday seems be the day that users did the most exercise lightly or moderately (accounting both the distance and minute on average)¶

However, based on the above visualizations and summary data, I can't really draw a solid conclusion. For now, I hyphothesize that users are probably most active on Saturdays.¶

What factors contribute to the highest calorie burns?¶

In [24]:
caloriesGrouped = dailyActivity.pivot_table(
    values = 'Calories',
    index = 'day_name'
)

caloriesGrouped
Out[24]:
Calories
day_name
Friday 2331.785714
Monday 2324.208333
Saturday 2354.967742
Sunday 2263.000000
Thursday 2199.571429
Tuesday 2356.013158
Wednesday 2302.620000

Based on the summary data, it would seem that people burn the least calorie on Thursdays on average, whereas Tuesday is has the most calories burned on average.¶

Let's analyze steps taken vs calories burned¶

In [25]:
fig_dims = (15, 10)
fig, ax = plt.subplots(figsize=fig_dims)

sns.set_style("darkgrid")
reg = sns.regplot(
    data = dailyActivity,
    x = 'TotalSteps',
    y = 'Calories',
    ax = ax,
    color = 'blue'
)

reg.set_title('Relationship Between Total Steps Taken and Calories Burned')
Out[25]:
Text(0.5, 1.0, 'Relationship Between Total Steps Taken and Calories Burned')

Based on this visualization, I can clearly see that there's a positive relationship between these total steps taken and calories burned. The more steps that were taken on a particular day, the more calories were burned. I guess this was probably very obvious as the more we move our bodies, we tend to burn more calories.¶

How does the data vary from weekends and weekdays?¶

How much calories do they burn between weekdays and weekends? How about the quality of their sleep?¶

I will now inner join between the sleepDay and dailyActivity dataset in order to analyze the users' sleeping behaviours.¶

In [26]:
# Ensuring proper formatting and creating a new column in order to assist in the joining process
sleepDay['SleepDay'] = pd.to_datetime(sleepDay['SleepDay'])
sleepDay['date'] = pd.to_datetime(sleepDay['SleepDay'])
dailyActivity['date'] = pd.to_datetime(dailyActivity['ActivityDate'])

# sleepDay[['TotalSleepRecords', 'TotalMinutesAsleep', 'TotalTimeInBed']]

#join sleep data with daily activity
joinedDf = dailyActivity.merge(sleepDay, on = ['Id', 'date'])
In [27]:
joinedDf.head()
Out[27]:
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories day_name weekend_weekday date SleepDay TotalSleepRecords TotalMinutesAsleep TotalTimeInBed
0 1503960366 2016-04-12 13162 8.50 8.50 0.0 1.88 0.55 6.06 0.0 25 13 328 728 1985 Tuesday Weekday 2016-04-12 2016-04-12 1 327 346
1 1503960366 2016-04-13 10735 6.97 6.97 0.0 1.57 0.69 4.71 0.0 21 19 217 776 1797 Wednesday Weekday 2016-04-13 2016-04-13 2 384 407
2 1503960366 2016-04-15 9762 6.28 6.28 0.0 2.14 1.26 2.83 0.0 29 34 209 726 1745 Friday Weekday 2016-04-15 2016-04-15 1 412 442
3 1503960366 2016-04-16 12669 8.16 8.16 0.0 2.71 0.41 5.04 0.0 36 10 221 773 1863 Saturday Weekend 2016-04-16 2016-04-16 2 340 367
4 1503960366 2016-04-17 9705 6.48 6.48 0.0 3.19 0.78 2.51 0.0 38 20 164 539 1728 Sunday Weekend 2016-04-17 2016-04-17 1 700 712

Calories burned between weekdays and weekends¶

In [28]:
dailyActivity.groupby('weekend_weekday')['Calories'].describe()
Out[28]:
count mean std min 25% 50% 75% max
weekend_weekday
Weekday 695.0 2301.516547 704.675507 0.0 1841.0 2159.0 2799.5 4900.0
Weekend 245.0 2309.546939 756.589860 0.0 1792.0 2096.0 2739.0 4552.0

Sleep quality between weekdays and weekends¶

In [29]:
joinedDf.groupby('weekend_weekday')[['TotalSleepRecords', 'TotalMinutesAsleep', 'TotalTimeInBed']].describe()
Out[29]:
TotalSleepRecords TotalMinutesAsleep TotalTimeInBed
count mean std min 25% 50% 75% max count mean std min 25% 50% 75% max count mean std min 25% 50% 75% max
weekend_weekday
Weekday 300.0 1.093333 0.313501 1.0 1.0 1.0 1.0 3.0 300.0 413.106667 103.459145 59.0 362.5 426.0 472.5 796.0 300.0 449.903333 108.342949 65.0 406.0 458.0 506.0 961.0
Weekend 113.0 1.185841 0.412931 1.0 1.0 1.0 1.0 3.0 113.0 436.353982 150.162288 58.0 361.0 465.0 530.0 775.0 113.0 481.831858 165.356428 61.0 402.0 501.0 575.0 961.0

Based on this data, it seems that the users have a much better sleep quality on average on the weekends. Also, it seems that they burned more calories on the weekends as well, which further confirms my earlier hypothesis that saturdays are the most popular day for users to engage in physical exercises.¶

Let's see the relationship between total sleep minutes and total time in bed subgrouped by the type of week¶

In [30]:
sleep = sns.relplot(
    data = joinedDf, 
    x = 'TotalMinutesAsleep',
    y = 'TotalTimeInBed',
    col = 'weekend_weekday',
    hue = 'TotalSleepRecords',
    height = 7
)

sleep.fig.suptitle('Total Minutes Asleep Vs Total Time in Bed',  y = 1.03)
sleep.set_titles("Week: {col_name}")
plt.show()

Both display a positive relationship. I guess it's common sense that the longer you stay in bed, the more likely you will fall into sleep. Let's analyze this in another way.¶

Based on a study here, healthy adults are suggested to sleep between 7-9 hours. Therefore, I will categorized those who slept below 7 as Bad Sleepers, those who slept beyond 9 hours Oversleepers, and those between 7 and 9 hours as normal.¶

In [31]:
# convert to proper minutes
sleepDay['TotalMinutesAsleep'] = pd.to_timedelta(sleepDay['TotalMinutesAsleep'], unit = 'm')
sleepDay['TotalMinutesAsleep'] = sleepDay['TotalMinutesAsleep'].astype('timedelta64[m]')

#convert to hours
sleepDay['TotalHoursAsleep'] = sleepDay['TotalMinutesAsleep'] / 60

# round to nearest hundreths 
sleepDay['TotalHoursAsleep'] = sleepDay['TotalHoursAsleep'].round(2)

#add sleep status
sleepDay['SleepStatus'] = np.where(sleepDay['TotalHoursAsleep'] < 7, 'Bad Sleepers', 
         (np.where( sleepDay['TotalHoursAsleep'] > 9, 'Oversleepers', 'Normal Sleepers'))
        ) 
#used ternary operator. link for future me's reference: https://stackoverflow.com/questions/39109045/numpy-where-with-multiple-conditions
#np.select article by dataquest: https://www.dataquest.io/blog/tutorial-add-column-pandas-dataframe-based-on-if-else-condition/

sleepDay['day_name'] = sleepDay['SleepDay'].dt.day_name()
sleepDay['weekend_weekday'] = np.where(sleepDay['SleepDay'].dt.dayofweek > 4, 'Weekend', 'Weekday')

sleepDay.head()

#Overview
fig_dims = (20, 15)
fig, ax = plt.subplots(figsize = fig_dims)
overview = sns.countplot(
    data = sleepDay,
    x = sleepDay['SleepStatus'],
    hue = 'weekend_weekday',
    ax = ax
)
overview.set_title('Overview Look of Sleep Statuses', fontsize = 20)
overview.set_xlabel('Sleep Status', fontsize = 20)
overview.set_ylabel('Total', fontsize = 20)
plt.xticks(size = 20)
plt.yticks(size = 20)

plt.show()

#by Day of the week
fig_dims = (20, 15)
fig, ax = plt.subplots(figsize = fig_dims)
overview2 = sns.countplot(
    data = sleepDay,
    x = sleepDay['SleepStatus'],
    hue = 'day_name',
    ax = ax
)
overview2.set_title('Overview Look of Sleep Statuses by Day of The Week', fontsize = 20)
overview2.set_xlabel('Sleep Status', fontsize = 20)
overview2.set_ylabel('Total', fontsize = 20)
plt.xticks(size = 20)
plt.yticks(size = 20)

plt.show()

Overall it seems that people tend to oversleep the most on a Sunday, and they often sleep below 7 hours the most on a Tuesday. It makes sense for many people to oversleep on a Sunday as most people won't need to work. However, as I've categorized those sleeping over 9 hours as Bad Sleepers, Bellabeat should make an effort to advice its users to avoid sleeping in on Sundays in order to not damage their health.¶

Also it seems that the users underslept the most on Tuesdays. Not sure why this is the case, I should try to investigate this.¶

Let's look back at my summary data based on the users' active status.¶

In [32]:
byMinute
Out[32]:
FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes VeryActiveMinutes
day_name
Friday 12.111111 204.198413 1000.309524 20.055556
Monday 14.000000 192.058333 1027.941667 23.108333
Saturday 15.201613 207.145161 964.282258 21.919355
Sunday 14.528926 173.975207 990.256198 19.983471
Thursday 11.959184 185.421769 961.993197 19.408163
Tuesday 14.335526 197.342105 1007.361842 22.953947
Wednesday 13.100000 189.853333 989.480000 20.780000

A quick glimpse back, the users are the most idle on Mondays, and other than their Very Active Minutes, the average for the rest seemed to have increased on Tuesday. However, there isn't much I could tell just based on these information as there's not much signicant difference. Maybe the heartrate dataset can help me?¶

Let's analyze the heartrate dataset¶

In [33]:
# There's 2,483,658 rows!
heartrateSeconds.shape
print(heartrateSeconds.head())
summary = round(heartrateSeconds['Value'].describe())
print('\n')
print(summary)
unique = heartrateSeconds['Id'].nunique() #there's only 14 unique uses
print('\nNumber of Users in This Dataset: ' + str(unique))

           Id                  Time  Value
0  2022484408  4/12/2016 7:21:00 AM     97
1  2022484408  4/12/2016 7:21:05 AM    102
2  2022484408  4/12/2016 7:21:10 AM    105
3  2022484408  4/12/2016 7:21:20 AM    103
4  2022484408  4/12/2016 7:21:25 AM    101


count    2483658.0
mean          77.0
std           19.0
min           36.0
25%           63.0
50%           73.0
75%           88.0
max          203.0
Name: Value, dtype: float64

Number of Users in This Dataset: 14

Since FitBit collects heart-rates in BPM, or Beats Per Minute, the Value column is therefore the BPM collected. According to FitBit's website, a typical resting heart-rate range is 60 to 100 beats per minute for adults. It also mentions that active people often have a lower resting heart rate, therefore drastic changes to this number can signify health issues and should be discussed with a doctor.¶

Moving on to the summary:¶

  1. The average heart rate is 77, which is inside the range of a typical resting heart-rate range.
  2. Seems like there's someone that had a maximun BPM of 203. That is quite worrisome. Bellabeat should consider implementing a system that allows their users to be notified if their BPM is outside the range of 60 to 100 and be checked out by a doctor.

Alright let's start with analyzing whether the heart-rate has anything to do with people undersleeping.¶

In [34]:
#Ensure proper time formatting
heartrateSeconds['Time'] = pd.to_datetime(heartrateSeconds['Time'], format = '%m/%d/%Y %I:%M:%S %p')

mean = heartrateSeconds.groupby(heartrateSeconds['Time'].dt.day_name())['Value'].mean()
mean
Out[34]:
Time
Friday       77.520836
Monday       77.454335
Saturday     79.973815
Sunday       75.925004
Thursday     77.035902
Tuesday      77.013723
Wednesday    76.451580
Name: Value, dtype: float64

Sadly, it doesn't seem like information on the heart-rate seem to be of help here. But the average seems to be highest on Saturdays, which I'd assume it's due to the fact that the users exercise the most on Saturdays. It doesn't seem I would be able to determine why some people undersleep with the information I currently have right now.¶

Let's look at the average heart-rate per minute for each hour of the day¶

In [35]:
mean = heartrateSeconds.groupby(heartrateSeconds['Time'].dt.hour)['Value'].mean()
fig_dims = (23, 13)
fig, ax = plt.subplots(figsize=fig_dims)

# create temporary dataframe
tempDf = pd.DataFrame(mean)

tempDf

t = sns.barplot(
    data = tempDf, 
    x = tempDf.index,
    y = 'Value'
)

t.set_xlabel('Time (Hour)', fontsize = 20)
t.set_ylabel('Average HeartRate', fontsize = 20)
t.set_title('Average Heart-Rate Per Minute Based on Time (Hour)', fontsize = 20)
Out[35]:
Text(0.5, 1.0, 'Average Heart-Rate Per Minute Based on Time (Hour)')

Average Heart-rate per minute is generally slower during the night, which is understandable since most people will be resting during the night. Heart-rate is high between 9am to 6pm, and starts to drop after 6pm. Heart rate also seems to be at its highest at 12pm and 6pm. Seems like users are somewhat tensed during these two distinct hours.¶

Let's try analyzing daily steps¶

In [36]:
#rejoining a new df as sleepDay was altered
rejoined = dailyActivity.merge(sleepDay, on = ['Id', 'date'])
rejoined.head()
Out[36]:
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories day_name_x weekend_weekday_x date SleepDay TotalSleepRecords TotalMinutesAsleep TotalTimeInBed TotalHoursAsleep SleepStatus day_name_y weekend_weekday_y
0 1503960366 2016-04-12 13162 8.50 8.50 0.0 1.88 0.55 6.06 0.0 25 13 328 728 1985 Tuesday Weekday 2016-04-12 2016-04-12 1 327.0 346 5.45 Bad Sleepers Tuesday Weekday
1 1503960366 2016-04-13 10735 6.97 6.97 0.0 1.57 0.69 4.71 0.0 21 19 217 776 1797 Wednesday Weekday 2016-04-13 2016-04-13 2 384.0 407 6.40 Bad Sleepers Wednesday Weekday
2 1503960366 2016-04-15 9762 6.28 6.28 0.0 2.14 1.26 2.83 0.0 29 34 209 726 1745 Friday Weekday 2016-04-15 2016-04-15 1 412.0 442 6.87 Bad Sleepers Friday Weekday
3 1503960366 2016-04-16 12669 8.16 8.16 0.0 2.71 0.41 5.04 0.0 36 10 221 773 1863 Saturday Weekend 2016-04-16 2016-04-16 2 340.0 367 5.67 Bad Sleepers Saturday Weekend
4 1503960366 2016-04-17 9705 6.48 6.48 0.0 3.19 0.78 2.51 0.0 38 20 164 539 1728 Sunday Weekend 2016-04-17 2016-04-17 1 700.0 712 11.67 Oversleepers Sunday Weekend
In [37]:
# Before I analyze with sleep status, let's examine just the total steps taken by weekdays

meanSteps = dailyActivity.groupby('day_name')['TotalSteps'].mean()

fig_dims = (23, 13)
fig, ax = plt.subplots(figsize=fig_dims)

# create temporary dataframe
tempDf = pd.DataFrame(meanSteps)
tempDf = tempDf.reset_index()

#sort by day
cats = ['Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday', 'Sunday']
tempDf['day_name'] = pd.Categorical(tempDf['day_name'], categories = cats)
tempDf = tempDf.sort_values('day_name')

print(tempDf)

t = sns.barplot(
    data = tempDf, 
    x = 'day_name',
    y = 'TotalSteps'
)

t.set_xlabel('Day of the Week', fontsize = 20)
t.set_ylabel('Total Steps', fontsize = 20)
t.set_title('Average Total Steps Taken by the Day of the Week', fontsize = 20)

    day_name   TotalSteps
1     Monday  7780.866667
5    Tuesday  8125.006579
6  Wednesday  7559.373333
4   Thursday  7405.836735
0     Friday  7448.230159
2   Saturday  8152.975806
3     Sunday  6933.231405
Out[37]:
Text(0.5, 1.0, 'Average Total Steps Taken by the Day of the Week')

On weekdays, Tuesday has the most steps taken, whereas on weekends it has the most steps taken on Saturdays. I hyphothesize that since Tuesday seems to be the most busiest day among the users, this may be the very reason why users are mostly undersleeping on Tuesdays.¶

Is there a relationship between how active they are and their sleep schedule?¶

Since most of these users are lightly active, let's see if there's any relationship between light active users and their sleeping behaviours.¶

In [38]:
lightlyActive = joinedDf[['LightlyActiveMinutes', 'TotalSleepRecords', 'TotalMinutesAsleep', 'TotalTimeInBed']]

lightlyActive.head()
Out[38]:
LightlyActiveMinutes TotalSleepRecords TotalMinutesAsleep TotalTimeInBed
0 328 1 327 346
1 217 2 384 407
2 209 1 412 442
3 221 2 340 367
4 164 1 700 712
In [39]:
fig_dims = (15, 10)
fig, ax = plt.subplots(figsize=fig_dims)

sns.set_style("darkgrid")
reg = sns.regplot(
    data = lightlyActive,
    x = 'LightlyActiveMinutes',
    y = 'TotalMinutesAsleep',
    ax = ax,
    color = 'blue'
)

reg.set_title('Relationship between Total Minutes of Sleep and Users Who are Light Active')
Out[39]:
Text(0.5, 1.0, 'Relationship between Total Minutes of Sleep and Users Who are Light Active')

Looks like there's no correlation between those who lightly active and the amount of sleep they have.¶

What about those who are mostly sedentary?¶

In [40]:
sedentary = joinedDf[['SedentaryMinutes', 'TotalSleepRecords', 'TotalMinutesAsleep', 'TotalTimeInBed']]

sedentary.head()
Out[40]:
SedentaryMinutes TotalSleepRecords TotalMinutesAsleep TotalTimeInBed
0 728 1 327 346
1 776 2 384 407
2 726 1 412 442
3 773 2 340 367
4 539 1 700 712
In [41]:
sns.set_style("darkgrid")
fig_dims = (15, 10)
fig, ax = plt.subplots(figsize=fig_dims)

s = sns.regplot(
    data = sedentary,
    x = 'SedentaryMinutes',
    y = 'TotalMinutesAsleep',
    ax = ax,
    color = 'blue'
)

s.set_title('Relationship between Total Minutes of Sleep and Users Who are Sedentary')
Out[41]:
Text(0.5, 1.0, 'Relationship between Total Minutes of Sleep and Users Who are Sedentary')

There's a negative correlation! The more idle the person is during the day, it seems that they would also be getting fewer sleep than those who are more active!¶

I recommend bellabeat to take note of this insight and maybe consider sending notifications to users who might have been idle throughout the day to get some light exercises so as to allow them to have a better sleep later in the day.¶

What About those who are fairly active and very active?¶

In [42]:
joinedDf.head()
Out[42]:
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories day_name weekend_weekday date SleepDay TotalSleepRecords TotalMinutesAsleep TotalTimeInBed
0 1503960366 2016-04-12 13162 8.50 8.50 0.0 1.88 0.55 6.06 0.0 25 13 328 728 1985 Tuesday Weekday 2016-04-12 2016-04-12 1 327 346
1 1503960366 2016-04-13 10735 6.97 6.97 0.0 1.57 0.69 4.71 0.0 21 19 217 776 1797 Wednesday Weekday 2016-04-13 2016-04-13 2 384 407
2 1503960366 2016-04-15 9762 6.28 6.28 0.0 2.14 1.26 2.83 0.0 29 34 209 726 1745 Friday Weekday 2016-04-15 2016-04-15 1 412 442
3 1503960366 2016-04-16 12669 8.16 8.16 0.0 2.71 0.41 5.04 0.0 36 10 221 773 1863 Saturday Weekend 2016-04-16 2016-04-16 2 340 367
4 1503960366 2016-04-17 9705 6.48 6.48 0.0 3.19 0.78 2.51 0.0 38 20 164 539 1728 Sunday Weekend 2016-04-17 2016-04-17 1 700 712
In [43]:
sns.set_style("darkgrid")
fig_dims = (15, 10)
fig, ax = plt.subplots(figsize=fig_dims)

s = sns.regplot(
    data = joinedDf,
    x = 'FairlyActiveMinutes',
    y = 'TotalMinutesAsleep',
    ax = ax,
    color = 'blue'
)

s.set_title('Relationship between Total Minutes of Sleep and Users Who are Fairly Active', fontsize = 20)
s.set_xlabel('Fairly Active Minutes', fontsize = 20)
s.set_ylabel('Total Minutes Asleep', fontsize = 20)
plt.show()

# Very Active 
sns.set_style("darkgrid")
fig_dims = (15, 10)
fig, ax = plt.subplots(figsize=fig_dims)

s = sns.regplot(
    data = joinedDf,
    x = 'VeryActiveMinutes',
    y = 'TotalMinutesAsleep',
    ax = ax,
    color = 'blue'
)

s.set_title('Relationship between Total Minutes of Sleep and Users Who are Very Active', fontsize = 20)
s.set_xlabel('Very Active Minutes', fontsize = 20)
s.set_ylabel('Total Minutes Asleep', fontsize = 20)
plt.show()

Although there's a slight negative trend for Fairly active users, it doesn't seem to be a very strong correlation but still worth mentioning. Also, those who are very active don't seem to have a correlation as well.¶

On what hour do users are the most active during the day?¶

In [44]:
#Ensuring proper formatting
hourlyIntensities['ActivityHour'] = pd.to_datetime(hourlyIntensities['ActivityHour'], format = '%m/%d/%Y %I:%M:%S %p')
In [45]:
#Group by the hours and find the average total intensity, which should therefore give only 24 rows in the end
mean = hourlyIntensities.groupby(hourlyIntensities['ActivityHour'].dt.time)['TotalIntensity'].mean()

sns.set_style("darkgrid")
fig_dims = (23, 13)
fig, ax = plt.subplots(figsize=fig_dims)

#create temporary dataframe
tempDf = pd.DataFrame(mean)

t = sns.barplot(
    data = tempDf, 
    x = tempDf.index,
    y = 'TotalIntensity'
)

t.set_xlabel('Time', fontsize = 20)
t.set_ylabel('Average Total Intensity', fontsize = 20)
t.set_title('Average Total Intensity Per Hour', fontsize = 20)
Out[45]:
Text(0.5, 1.0, 'Average Total Intensity Per Hour')

With this visualization I can see that:¶

  • Users are most active between 5pm to 7pm. I speculate that this is because users tend to finish work around these hours, and right after they start engaging in physical activities such as going to the gym or a jog. Therefore, I recommend Bellabeat to notify users to do some exercises around this time frame.
  • Users start to get active starting from 5am and rest on 9pm.

On what day of the week are users the most active during the day?¶

In [46]:
meanDays = hourlyIntensities.groupby(hourlyIntensities['ActivityHour'].dt.day_name())['TotalIntensity'].mean()

fig_dims = (23, 13)
fig, ax = plt.subplots(figsize=fig_dims)

# create temporary dataframe
tempDf = pd.DataFrame(meanDays)
tempDf = tempDf.reset_index()

tempDf.rename(columns = {"ActivityHour": "Day"}, inplace = True)

#sort by day
cats = ['Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday', 'Sunday']
tempDf['Day'] = pd.Categorical(tempDf['Day'], categories = cats)
tempDf = tempDf.sort_values('Day')

t= sns.barplot(
    x = 'Day',
    y = 'TotalIntensity',
    data = tempDf
)

t.set_xlabel('Day', fontsize = 20)
t.set_ylabel('Average Total Intensity', fontsize = 20)
t.set_title('Average Total Intensity Vs Day', fontsize = 20)
Out[46]:
Text(0.5, 1.0, 'Average Total Intensity Vs Day')

With the above visualization, I can see that Sunday is the least intense among the users and that Saturday is the most intense. Also, on weekdays Tuesdays seem to be the most busiest in terms of both average total intensity and total steps taken. With this much evidence, I can safely conclude that Saturdays are the most active day on weekends and Tuesdays the most busiest on weekdays.¶

Summary of my findings:¶

After analyzing the data, I have found some insightful information that may be of interest for Bellabeat. As a reminder, Bellabeat is a high-tech company that manufactures health-focused smart products for women. Here are some business marketing recommendations that I can give to Bellabeat:

  1. I suggest Bellabeat to give notifications to their users who had been idle during the day to consider doing some light exercises. And with my previous finding that shows that being sedentary is associating to lack of sleep, it shows that the more idle you are, the lesser the quality of one's sleep may be. Therefore, this reason may serve as a good incentive to encourage users to be more active as this may allow them to sleep better when it's time for bed.

  2. I recommend Bellabeat to notify/remind users to do some exercises between 5pm and 7pm as this is when most users finish work and start exercising based on the data. This is also a great opportunity to encourage those who might have a goal to lose weight to start exercising on this time frame.

  3. Bellabeat should consider notifying users the importance of at least getting more than 8000 steps per day as this will benefit the users in the long run. This is based on a research by CDC that taking 8,000 steps per day was associated with a 51% lower risk for all-cause mortality (or death from all causes). Therefore, it is highly recommended for Bellabeat to consider this advice into consideration for their next marketing strategy.

  4. As users tend exercise more on Saturdays, Bellabeat can use this information to remind users to do some exercise on Saturdays as they are more likely to be more motivated. Also, there seems to be a noticable drop in activity when it is Sunday. Therefore, I recommend Bellabeat to encourage users to be more proactive during the Sundays as, based on the data, users tend to be a bit more laidback on Sundays. Users also seem to oversleep the most on Sunday, therefore Bellabeat should consider advising their users to be mindful of this as oversleeping is associated with risk of diabetes, heart disease, stroke, and death (source)

  5. Bellabeat should notify users when their heart-rate detected to be outside the range of 60-100 BPM as this may signify a potential underlying health condition that the user may not be aware of.

  6. By combining the fact that Tuesday has the most steps taken and that average total intensity is the highest, this might be the very reason why the users on this dataset are undersleeping the most on Tuesday. Bellabeat should consider advising their users to exercise meditation so as to calm themselves on stressful days.