Rayen Feng

Logo

Hi, I'm Rayen I’m an aspiring data scientist. I've worked on a couple of person projects and this is my porfolio. You’re welcome to look around.



Check out my Blog!

Becoming top a top 3% chess player
Learning Kanji in 3 months
My Iceland Itenerary

Bella Beat Case Study

Analysis performed by: Rayen Feng
Date: June 27, 2022
As part of Google Data Analytics Certificate

Table of Contents

Introduction

Bellabeat is a high-tech manufacturer of heath products for women. By analyzing their smart products and using data-driven decision making can help discover and drive new opportunities for the company.

In this analysis, one of Bellabeat’s products will be chosen to undergo analysis using their smart device data to gain insight on how consumers are using smart devices. These insights will help guilde the marketing strategy for the company. High level reccomendations will be made to infrom Bellabeat’s marketing stategy. For this, exisiting trends in user activity using Fitbit data will be used.

The analysis hopes to seek to answer the following questions.

  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?

  4. A clear summary of the business task
  5. A description of all data sources used
  6. Documentation of any cleaning or manipulation of data
  7. A summary of your analysis
  8. Supporting visualizations and key findings
  9. Your top high-level content recommendations based on your analysis

Data Preperation and processing

As mentioned in the introduction, the following analysis will use the Kaggle dataset which contains personal fitness tracker from thirty fitbit users. Thirty eligible Fitbit users consented to the submission of personal tracker data, including minute-level output for physical activity, heart rate, and sleep monitoring. It includes information about daily activity, steps, and heart rate that can be used to explore users’ habits. For this analysis, python and the pandas dataframe package will be used for data cleaning, manipulation and analysis. Additionally, packages such as matplotlib and seaborn will be used to visualize and present the data.

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import datetime
pd.options.mode.chained_assignment = None  # default='warn'

import os
for dirname, _, filenames in os.walk('/kaggle/input'):
    for filename in filenames:
        print(os.path.join(dirname, filename))


/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteCaloriesNarrow_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/weightLogInfo_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/sleepDay_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/dailyIntensities_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteIntensitiesWide_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteMETsNarrow_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/dailyCalories_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/hourlyCalories_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/heartrate_seconds_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteSleep_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/hourlyIntensities_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/hourlySteps_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteStepsNarrow_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/dailySteps_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteStepsWide_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteIntensitiesNarrow_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteCaloriesWide_merged.csv
/kaggle/input/fitbit/Fitabase Data 4.12.16-5.12.16/dailyActivity_merged.csv

Importing data

After previewing the data, a few datasets were selected and imported to investigate further.

activity_daily = pd.read_csv ('../input/fitbit/Fitabase Data 4.12.16-5.12.16/dailyActivity_merged.csv')
heart_rate = pd.read_csv ('../input/fitbit/Fitabase Data 4.12.16-5.12.16/heartrate_seconds_merged.csv')
sleepDay_merged = pd.read_csv ('../input/fitbit/Fitabase Data 4.12.16-5.12.16/sleepDay_merged.csv')
minuteSleep_merged = pd.read_csv ('../input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteSleep_merged.csv')
minuteMETsNarrow_merged = pd.read_csv ('../input/fitbit/Fitabase Data 4.12.16-5.12.16/minuteMETsNarrow_merged.csv')

The first step will be to preview the main dataset, which is the daily_activity_merged dataset. This dataset contains the activity data of all 33 participlants and the breakdown for their activity for 30 days from April 12, 2016 to May 12, 2016. Looking at the rows shown below, the dataset contains the breakdown of the type of activity intensity, the logged distances and durations of summarized daily.

display(activity_daily)
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories
0 1503960366 4/12/2016 13162 8.500000 8.500000 0.0 1.88 0.55 6.06 0.00 25 13 328 728 1985
1 1503960366 4/13/2016 10735 6.970000 6.970000 0.0 1.57 0.69 4.71 0.00 21 19 217 776 1797
2 1503960366 4/14/2016 10460 6.740000 6.740000 0.0 2.44 0.40 3.91 0.00 30 11 181 1218 1776
3 1503960366 4/15/2016 9762 6.280000 6.280000 0.0 2.14 1.26 2.83 0.00 29 34 209 726 1745
4 1503960366 4/16/2016 12669 8.160000 8.160000 0.0 2.71 0.41 5.04 0.00 36 10 221 773 1863
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
935 8877689391 5/8/2016 10686 8.110000 8.110000 0.0 1.08 0.20 6.80 0.00 17 4 245 1174 2847
936 8877689391 5/9/2016 20226 18.250000 18.250000 0.0 11.10 0.80 6.24 0.05 73 19 217 1131 3710
937 8877689391 5/10/2016 10733 8.150000 8.150000 0.0 1.35 0.46 6.28 0.00 18 11 224 1187 2832
938 8877689391 5/11/2016 21420 19.559999 19.559999 0.0 13.22 0.41 5.89 0.00 88 12 213 1127 3832
939 8877689391 5/12/2016 8064 6.120000 6.120000 0.0 1.82 0.04 4.25 0.00 23 1 137 770 1849

940 rows × 15 columns

display(activity_daily.max())
activity_daily.min()

Id                          8877689391
ActivityDate                  5/9/2016
TotalSteps                       36019
TotalDistance                28.030001
TrackerDistance              28.030001
LoggedActivitiesDistance      4.942142
VeryActiveDistance               21.92
ModeratelyActiveDistance          6.48
LightActiveDistance              10.71
SedentaryActiveDistance           0.11
VeryActiveMinutes                  210
FairlyActiveMinutes                143
LightlyActiveMinutes               518
SedentaryMinutes                  1440
Calories                          4900
dtype: object





Id                          1503960366
ActivityDate                 4/12/2016
TotalSteps                           0
TotalDistance                      0.0
TrackerDistance                    0.0
LoggedActivitiesDistance           0.0
VeryActiveDistance                 0.0
ModeratelyActiveDistance           0.0
LightActiveDistance                0.0
SedentaryActiveDistance            0.0
VeryActiveMinutes                    0
FairlyActiveMinutes                  0
LightlyActiveMinutes                 0
SedentaryMinutes                     0
Calories                             0
dtype: object

A quick conversion of the activity date to datetime format can be done to clean the dataset.

activity_daily['ActivityDate'] = pd.to_datetime(activity_daily['ActivityDate'])

To check the general validity and accuracy of this data, a quick check can be performed by plotting the relationship of two variables. It’s known that generally, the more distance traveled or the more steps walked, the more calories are burned. By plotting these two variables in a scatterplot, a general upward trend of these two varibles is observed, which proves the validity of this data.

sns.scatterplot(
    x='TotalDistance',
    y='Calories',
    data = activity_daily,
)

plt.title("Distance vs. Calories Burned")
plt.xlabel("Distance Traveled km")
plt.ylabel("Calories Burned")

plt.show()

sns.scatterplot(
    x='TotalSteps',
    y='Calories',
    data = activity_daily,
)

plt.title("Steps vs. Calories Burned")
plt.xlabel("Steps")
plt.ylabel("Calories Burned")

plt.show()

png

png

Preparing User data

The activity data can be grouped and then grouped by their user ID to find general average information of each user. Additionally, a quick check can be made to check that the total users in this dataset is 33.

activity_daily_indv = activity_daily.groupby(['Id']).mean()
display(activity_daily_indv.sort_values(by = ["Calories"], ascending = False).head(n=5))
print(len(activity_daily_indv))
TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories
Id
8378563200 8717.709677 6.913548 6.913548 1.116158 2.503548 0.519032 3.889355 0.000000 58.677419 10.258065 156.096774 716.129032 3436.580645
8877689391 16040.032258 13.212903 13.212903 0.000000 6.637419 0.337742 6.188710 0.005161 66.064516 9.935484 234.709677 1112.870968 3420.258065
5577150313 8304.433333 6.213333 6.213333 0.000000 3.113667 0.658000 2.428000 0.000000 87.333333 29.833333 147.933333 754.433333 3359.633333
4388161847 10813.935484 8.393226 8.393226 0.000000 1.719355 0.901935 5.396129 0.000000 23.161290 20.354839 229.354839 836.677419 3093.870968
4702921684 8572.064516 6.955161 6.955161 0.000000 0.417419 1.304839 5.225484 0.000000 5.129032 26.032258 237.483871 766.419355 2965.548387 
33

One last consideration to prepare this data is the participation rate of the users. Users who didn’t wear the watch as much might have a negative impact on the data, especially when considering what the participants use the watch for. To account for this, the users that participated less will have to be filtered out. This threshold will be set at a 80% participation rate.

By splitting these users into two groups, there is flexibility to see trends between the groups as well.

# create a function that shows the activity data for one user for a specific dataframe 

def show_user_data(df,user_id):
    
    t = df[df.Id == user_id] 
    return(t)

# Finding participants who took than 50 total steps per day

unique_ids = activity_daily['Id'].unique()

days_skip_arr = [] 

for i in unique_ids: 
    person_data = show_user_data(activity_daily,i)
    day_skipped = person_data[person_data.TotalSteps <= 50] 
    days_skipped = len(day_skipped)
    days_skip_arr.append(days_skipped)

    
day_skip_by_id = {'Id': unique_ids, 'days_skipped':days_skip_arr }
day_skip_by_ids= pd.DataFrame(data=day_skip_by_id)


# seperating the participants ids to 80% active and inactive 

day_skip_by_ids_sorted = day_skip_by_ids.sort_values(by = ['days_skipped'], ascending = False).reset_index(drop = True)

ids_80_percent_inactive = day_skip_by_ids_sorted.iloc[:5]
ids_80_percent_active = day_skip_by_ids_sorted.iloc[5:]

# splitting the activity_data between active and inactive users 

active_users = activity_daily[activity_daily.Id.isin(ids_80_percent_active.Id)] 
inactive_users = activity_daily[activity_daily.Id.isin(ids_80_percent_inactive.Id)] 

display(active_users)
display(inactive_users)
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories
0 1503960366 2016-04-12 13162 8.500000 8.500000 0.0 1.88 0.55 6.06 0.00 25 13 328 728 1985
1 1503960366 2016-04-13 10735 6.970000 6.970000 0.0 1.57 0.69 4.71 0.00 21 19 217 776 1797
2 1503960366 2016-04-14 10460 6.740000 6.740000 0.0 2.44 0.40 3.91 0.00 30 11 181 1218 1776
3 1503960366 2016-04-15 9762 6.280000 6.280000 0.0 2.14 1.26 2.83 0.00 29 34 209 726 1745
4 1503960366 2016-04-16 12669 8.160000 8.160000 0.0 2.71 0.41 5.04 0.00 36 10 221 773 1863
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
935 8877689391 2016-05-08 10686 8.110000 8.110000 0.0 1.08 0.20 6.80 0.00 17 4 245 1174 2847
936 8877689391 2016-05-09 20226 18.250000 18.250000 0.0 11.10 0.80 6.24 0.05 73 19 217 1131 3710
937 8877689391 2016-05-10 10733 8.150000 8.150000 0.0 1.35 0.46 6.28 0.00 18 11 224 1187 2832
938 8877689391 2016-05-11 21420 19.559999 19.559999 0.0 13.22 0.41 5.89 0.00 88 12 213 1127 3832
939 8877689391 2016-05-12 8064 6.120000 6.120000 0.0 1.82 0.04 4.25 0.00 23 1 137 770 1849

792 rows × 15 columns

Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories
92 1844505072 2016-04-12 6697 4.43 4.43 0.0 0.0 0.0 4.43 0.0 0 0 339 1101 2030
93 1844505072 2016-04-13 4929 3.26 3.26 0.0 0.0 0.0 3.26 0.0 0 0 248 1192 1860
94 1844505072 2016-04-14 7937 5.25 5.25 0.0 0.0 0.0 5.23 0.0 0 0 373 843 2130
95 1844505072 2016-04-15 3844 2.54 2.54 0.0 0.0 0.0 2.54 0.0 0 0 176 527 1725
96 1844505072 2016-04-16 3414 2.26 2.26 0.0 0.0 0.0 2.26 0.0 0 0 147 1293 1657
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
904 8792009665 2016-05-06 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688
905 8792009665 2016-05-07 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688
906 8792009665 2016-05-08 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688
907 8792009665 2016-05-09 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688
908 8792009665 2016-05-10 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 48 57

148 rows × 15 columns

Analysis of specific questions

To answer the question about how participants use smartwatches, a few trends need to be analyzed further.

  1. Do people have a preference towards using the watch as a sleep tracker or as an activity tracker?
  2. What kinds of activities does a user do while using the watch?
  3. How long do they have their watch on for the day?

To calculate the type of activity that the user does most frequently while wearing the watch, the time from each activty can be summed.

sum_col_inact = inactive_users["VeryActiveMinutes"] + inactive_users["FairlyActiveMinutes"] + inactive_users["LightlyActiveMinutes"] + inactive_users["SedentaryMinutes"] 
inactive_users["Total_Minutes"] = sum_col_inact

sum_col_act = active_users["VeryActiveMinutes"] + active_users["FairlyActiveMinutes"] + active_users["LightlyActiveMinutes"] + active_users["SedentaryMinutes"] 
active_users["Total_Minutes"] = sum_col_act

inactive_users
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories Total_Minutes
92 1844505072 2016-04-12 6697 4.43 4.43 0.0 0.0 0.0 4.43 0.0 0 0 339 1101 2030 1440
93 1844505072 2016-04-13 4929 3.26 3.26 0.0 0.0 0.0 3.26 0.0 0 0 248 1192 1860 1440
94 1844505072 2016-04-14 7937 5.25 5.25 0.0 0.0 0.0 5.23 0.0 0 0 373 843 2130 1216
95 1844505072 2016-04-15 3844 2.54 2.54 0.0 0.0 0.0 2.54 0.0 0 0 176 527 1725 703
96 1844505072 2016-04-16 3414 2.26 2.26 0.0 0.0 0.0 2.26 0.0 0 0 147 1293 1657 1440
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
904 8792009665 2016-05-06 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688 1440
905 8792009665 2016-05-07 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688 1440
906 8792009665 2016-05-08 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688 1440
907 8792009665 2016-05-09 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 1440 1688 1440
908 8792009665 2016-05-10 0 0.00 0.00 0.0 0.0 0.0 0.00 0.0 0 0 0 48 57 48

148 rows × 16 columns

display(active_users)

total_VAM = sum(active_users["VeryActiveMinutes"])
total_FAM = sum(active_users["FairlyActiveMinutes"])
toatl_LAM = sum(active_users["LightlyActiveMinutes"])
total_SEM = sum(active_users["SedentaryMinutes"])
total_min = sum(sum_col_act)

y = np.array([total_VAM,total_FAM,toatl_LAM,total_SEM])

mylabels = ["VeryActiveMinutes", "FairlyActiveMinutes", "LightlyActiveMinutes", "SedentaryMinutes"]
colors = sns.color_palette('pastel')
plt.pie(y, labels = mylabels, colors = colors)


plt.title("Time spent on type of Activity")
plt.show()
Id ActivityDate TotalSteps TotalDistance TrackerDistance LoggedActivitiesDistance VeryActiveDistance ModeratelyActiveDistance LightActiveDistance SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes Calories Total_Minutes
0 1503960366 2016-04-12 13162 8.500000 8.500000 0.0 1.88 0.55 6.06 0.00 25 13 328 728 1985 1094
1 1503960366 2016-04-13 10735 6.970000 6.970000 0.0 1.57 0.69 4.71 0.00 21 19 217 776 1797 1033
2 1503960366 2016-04-14 10460 6.740000 6.740000 0.0 2.44 0.40 3.91 0.00 30 11 181 1218 1776 1440
3 1503960366 2016-04-15 9762 6.280000 6.280000 0.0 2.14 1.26 2.83 0.00 29 34 209 726 1745 998
4 1503960366 2016-04-16 12669 8.160000 8.160000 0.0 2.71 0.41 5.04 0.00 36 10 221 773 1863 1040
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
935 8877689391 2016-05-08 10686 8.110000 8.110000 0.0 1.08 0.20 6.80 0.00 17 4 245 1174 2847 1440
936 8877689391 2016-05-09 20226 18.250000 18.250000 0.0 11.10 0.80 6.24 0.05 73 19 217 1131 3710 1440
937 8877689391 2016-05-10 10733 8.150000 8.150000 0.0 1.35 0.46 6.28 0.00 18 11 224 1187 2832 1440
938 8877689391 2016-05-11 21420 19.559999 19.559999 0.0 13.22 0.41 5.89 0.00 88 12 213 1127 3832 1440
939 8877689391 2016-05-12 8064 6.120000 6.120000 0.0 1.82 0.04 4.25 0.00 23 1 137 770 1849 931

792 rows × 16 columns

png

Key Findings of user activity

From this analysis, we can see the majority of users are engaged in sedentrary activity,followed by Light Activity. Only a small amount of time is spend performing moderate to intense activity

This data can show:

  1. It may be better to advertise smartwatch products to the general public as a tool that tracks general heath and fitness instead of a precise measurment tool for hardcore athelets.
  2. On the same note, this means that the advertisements should focus more on emphasizing the ease of usability and simplicity for tracking basic features such as heart rate and such, rather than marketing the wide use of advaced and complex features.
  3. Since most users spend times wearing the watch while inactive,and in a more causual setting, it might be good to focus efforts on emphasizing the asthetics of the watch.

Sleep vs. activity tracking usage

For this, a quick comparison between the number of users who logged into the sleep data vs the total number of users will be compared. To analyze this, the sleepDay_merged will be used to count the users who tracked their sleep.

display(sleepDay_merged)
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
... ... ... ... ... ...
408 8792009665 4/30/2016 12:00:00 AM 1 343 360
409 8792009665 5/1/2016 12:00:00 AM 1 503 527
410 8792009665 5/2/2016 12:00:00 AM 1 415 423
411 8792009665 5/3/2016 12:00:00 AM 1 516 545
412 8792009665 5/4/2016 12:00:00 AM 1 439 463

413 rows × 5 columns

# getting the number of users using the sleep tracking device / 30 users
unique_ids_sleep = sleepDay_merged['Id'].unique()
len(unique_ids_sleep)

unique_ids_sleep_mins = minuteSleep_merged['Id'].unique()
print("number of users who wear a watch to sleep: " + str(len(unique_ids_sleep_mins)))

number of users who wear a watch to sleep: 24

Now, it’s known that the number of users who wear the watch is less than the total number participants. However, this is not yet conclusive to whether or not users prefer one feature or the other. In order to eliminate the possibility of users who prefer not to wear the watch regardless, a additional process must be taken to exclude those users from this case.

active_user_id = active_users['Id'].unique()
unique_ids_sleep_mins

count_of_users = set(active_user_id).intersection(unique_ids_sleep_mins)
print("active users: " + str(len(active_user_id)))
print("active users who wear the watch to sleep: " + str(len(count_of_users)))

active users: 28
active users who wear the watch to sleep: 19

this means that 19 out of the 28 active users actually track their sleep data. This might mean that they are uncomforatable wearing this to bed. It might be better for bellafit to focus on marketing targeting the activity. Or, this can be a good oprortunity to fill in the gap and advertise it as a comfortable watch that can also track the sleep data.

Watch wear time relationships

there are three methods to track the watch time

  1. Through the activity data and counting the total active minutes
  2. Through the heart monitor

To begin with, a quick look at the watch useage over time can be analyzed to see if there is correlation between the time into the project and the watch wear time.

avg_active_mins_daily_active = active_users.groupby(['ActivityDate']).mean()
avg_active_mins_daily_inactive = inactive_users.groupby(['ActivityDate']).mean()

avg_active_mins_daily_active.plot(y = "Total_Minutes", figsize = (8,6))
plt.title("Day vs Minutes active of consistant watch users")
plt.xlabel("Date")
plt.ylabel("Minutes of Activity")

Text(0, 0.5, 'Minutes of Activity')

png

avg_active_mins_daily_inactive.plot(y = "Total_Minutes", figsize = (8,6))

plt.title("Day vs Minutes active of inconsistant watch users")
plt.xlabel("Date")
plt.ylabel("Minutes of Activity")


Text(0, 0.5, 'Minutes of Activity')

png

From this analysis, it can be noted that there is no direct correlation between the time and the duration of usage. However, it is noted that there might be a possibility that the participants had a lower average usage of the watch towards the end of the periods.

Finding average wear time

if the start day and wearing the watch is 7AM for example and the end time is 9:00PM assuming that when there’s no meter it means that the watch is off the person. By knowing what time they like to take it on and off shows if they like to track sleep or not.

display(heart_rate)
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
... ... ... ...
2483653 8877689391 5/12/2016 2:43:53 PM 57
2483654 8877689391 5/12/2016 2:43:58 PM 56
2483655 8877689391 5/12/2016 2:44:03 PM 55
2483656 8877689391 5/12/2016 2:44:18 PM 55
2483657 8877689391 5/12/2016 2:44:28 PM 56

2483658 rows × 3 columns

# converting to datetime format
heart_rate['Time'] = pd.to_datetime(heart_rate['Time'], format= "%m/%d/%Y  %I:%M:%S %p")

#creating new date and time columns
heart_rate['new_date'] = [d.date() for d in heart_rate['Time']]
heart_rate['new_time'] = [d.time() for d in heart_rate['Time']]

heart_rate
Id Time Value new_date new_time
0 2022484408 2016-04-12 07:21:00 97 2016-04-12 07:21:00
1 2022484408 2016-04-12 07:21:05 102 2016-04-12 07:21:05
2 2022484408 2016-04-12 07:21:10 105 2016-04-12 07:21:10
3 2022484408 2016-04-12 07:21:20 103 2016-04-12 07:21:20
4 2022484408 2016-04-12 07:21:25 101 2016-04-12 07:21:25
... ... ... ... ... ...
2483653 8877689391 2016-05-12 14:43:53 57 2016-05-12 14:43:53
2483654 8877689391 2016-05-12 14:43:58 56 2016-05-12 14:43:58
2483655 8877689391 2016-05-12 14:44:03 55 2016-05-12 14:44:03
2483656 8877689391 2016-05-12 14:44:18 55 2016-05-12 14:44:18
2483657 8877689391 2016-05-12 14:44:28 56 2016-05-12 14:44:28

2483658 rows × 5 columns

unique_dates= heart_rate['new_date'].unique()
unique_ids_heart = heart_rate['Id'].unique()

count_of_users_heart = set(active_user_id).intersection(unique_ids_heart)
print(str(len(count_of_users_heart)) + " out of " + str(len(active_user_id)))


11 out of 28

It’s worth noting that there is less than half of the people using tracking heartrate out of 28 active watch users. However, assuming that there is a indicator of a heartrate when the user keeps the watch on and no meter when the watch is off, this dataset would be a good measure to how long a user wears the watch for.

# calculates the watch worn time based on the difference between start and end times of given user and date

def watch_worn_time(df, user_id, date):
    
    user = df[df.Id == user_id]
    dates = user[user.new_date == date]
    start_time = dates['Time'].iloc[0]
    end_time =  dates['Time'].iloc[-1]
    delta_times =  end_time - start_time
    
    #print(start_time)
    #print(end_time)
    return delta_times / datetime.timedelta(hours=1)

worn_arr = [] 
date_arr = [] 


for j in unique_ids_heart: 
    # filter for the user 
    user_info = heart_rate[heart_rate.Id == j]
    # get unique dates from user
    unique_dates= user_info['new_date'].unique()

    # for every date the user logged in, calculate the time diff for each day. 
    for i in unique_dates:
        timediff = watch_worn_time(heart_rate,j,i)
        worn_arr.append(timediff)
        date_arr.append(i)
        
    

# date vs hours worn 

user_hours_worn = pd.DataFrame()

user_hours_worn['date'] = date_arr
user_hours_worn['hours_worn'] = worn_arr



user_hours_worn
date hours_worn
0 2016-04-12 12.512500
1 2016-04-13 12.880556
2 2016-04-14 12.827778
3 2016-04-15 12.077778
4 2016-04-16 13.173611
... ... ...
329 2016-05-08 14.091944
330 2016-05-09 15.506667
331 2016-05-10 14.584167
332 2016-05-11 15.582778
333 2016-05-12 7.967500

334 rows × 2 columns

worn_time_mean = user_hours_worn.groupby(['date']).mean()
#average time worn for each day
worn_time_mean.plot(figsize = (18,6))


plt.title("Date vs Average Time worn")
plt.xlabel("Date")
plt.ylabel("Hours worn")

Text(0, 0.5, 'Hours worn')

png

it can be seen that most users wear the watch anywhere from 16 to 22 hours per day. This is probably to track the sleep usage along with. With users wearing for the watch for long periods of time, it might be a good idea to empahsize the watches comfort.

Start and end time analysis

Next, if there is a concrete time that the watch is worn, it can be another indicator towards if people are inclined to wear the watch and sleep, or do people take it off and re-equip it during the day.

def watch_start_end_times(df, user_id, date):
    
    user = df[df.Id == user_id]
    dates = user[user.new_date == date]
    start_time = dates['Time'].iloc[0]
    end_time =  dates['Time'].iloc[-1]

    return start_time,end_time

# a second function that's slower was created to verify the total time. This runtime is slower but more accurate. Since the results from 
# both were similar, the first function can be used. 

# def watch_worn_time_2(df, user_id, date):
    
#     user = df[df.Id == user_id]
#     dates = user[user.new_date == date]
#     display()
#     accumulative_time = datetime.timedelta(0)
    
#     for i in range(0,len(dates)-1):
        
#         delta_times = dates['Time'].iloc[i+1] - dates['Time'].iloc[i]
#         accumulative_time += delta_times
        
#     return accumulative_time

# watch_worn_time_2(heart_rate,unique_ids_heart[0],unique_dates[16])

start_times = [] 
end_times = [] 


for j in unique_ids_heart: 
    # filter for the user 
    user_info = heart_rate[heart_rate.Id == j]
    # get unique dates from user
    unique_dates= user_info['new_date'].unique()

    # for every date the user logged in, calculate the time diff for each day. 
    for i in unique_dates:
        st_time, end_time = watch_start_end_times(heart_rate,j,i)
        start_times.append(st_time)
        end_times.append(end_time)

len(start_times)
len(end_times)

334

user_start_end_times = pd.DataFrame()

user_start_end_times['start_times'] = start_times
user_start_end_times['end_times'] = end_times

user_start_end_times
start_times end_times
0 2016-04-12 07:21:00 2016-04-12 19:51:45
1 2016-04-13 07:01:00 2016-04-13 19:53:50
2 2016-04-14 07:49:10 2016-04-14 20:38:50
3 2016-04-15 07:37:05 2016-04-15 19:41:45
4 2016-04-16 07:53:30 2016-04-16 21:03:55
... ... ...
329 2016-05-08 07:38:24 2016-05-08 21:43:55
330 2016-05-09 06:42:34 2016-05-09 22:12:58
331 2016-05-10 06:55:26 2016-05-10 21:30:29
332 2016-05-11 06:56:53 2016-05-11 22:31:51
333 2016-05-12 06:46:25 2016-05-12 14:44:28

334 rows × 2 columns

user_start_end_times['start_times_new'] = [d.time() for d in user_start_end_times['start_times']]
user_start_end_times['end_times_new'] = [d.time() for d in user_start_end_times['end_times']]

user_start_end_times
start_times end_times start_times_new end_times_new
0 2016-04-12 07:21:00 2016-04-12 19:51:45 07:21:00 19:51:45
1 2016-04-13 07:01:00 2016-04-13 19:53:50 07:01:00 19:53:50
2 2016-04-14 07:49:10 2016-04-14 20:38:50 07:49:10 20:38:50
3 2016-04-15 07:37:05 2016-04-15 19:41:45 07:37:05 19:41:45
4 2016-04-16 07:53:30 2016-04-16 21:03:55 07:53:30 21:03:55
... ... ... ... ...
329 2016-05-08 07:38:24 2016-05-08 21:43:55 07:38:24 21:43:55
330 2016-05-09 06:42:34 2016-05-09 22:12:58 06:42:34 22:12:58
331 2016-05-10 06:55:26 2016-05-10 21:30:29 06:55:26 21:30:29
332 2016-05-11 06:56:53 2016-05-11 22:31:51 06:56:53 22:31:51
333 2016-05-12 06:46:25 2016-05-12 14:44:28 06:46:25 14:44:28

334 rows × 4 columns

start_times_seconds = []
end_times_seconds = []

for t in user_start_end_times['start_times_new']:
    
    seconds = (t.hour * 60 + t.minute) * 60 + t.second
    start_times_seconds.append(seconds)

user_start_end_times['start_times_seconds'] = start_times_seconds

for t in user_start_end_times['end_times_new']:
    
    seconds = (t.hour * 60 + t.minute) * 60 + t.second
    end_times_seconds.append(seconds)

user_start_end_times['end_times_seconds'] = end_times_seconds

user_start_end_times
start_times end_times start_times_new end_times_new start_times_seconds end_times_seconds
0 2016-04-12 07:21:00 2016-04-12 19:51:45 07:21:00 19:51:45 26460 71505
1 2016-04-13 07:01:00 2016-04-13 19:53:50 07:01:00 19:53:50 25260 71630
2 2016-04-14 07:49:10 2016-04-14 20:38:50 07:49:10 20:38:50 28150 74330
3 2016-04-15 07:37:05 2016-04-15 19:41:45 07:37:05 19:41:45 27425 70905
4 2016-04-16 07:53:30 2016-04-16 21:03:55 07:53:30 21:03:55 28410 75835
... ... ... ... ... ... ...
329 2016-05-08 07:38:24 2016-05-08 21:43:55 07:38:24 21:43:55 27504 78235
330 2016-05-09 06:42:34 2016-05-09 22:12:58 06:42:34 22:12:58 24154 79978
331 2016-05-10 06:55:26 2016-05-10 21:30:29 06:55:26 21:30:29 24926 77429
332 2016-05-11 06:56:53 2016-05-11 22:31:51 06:56:53 22:31:51 25013 81111
333 2016-05-12 06:46:25 2016-05-12 14:44:28 06:46:25 14:44:28 24385 53068

334 rows × 6 columns

average_start_wear = user_start_end_times['start_times_seconds'].mean()
average_end_wear = user_start_end_times['end_times_seconds'].mean()

print(str(datetime.timedelta(seconds=average_start_wear)))
print(str(datetime.timedelta(seconds=average_end_wear)))

3:18:03
22:00:21.365269

It can’t be confirmed that users like to take off their watches to sleep. However, it slightly supports the notion of less users wearing it to bed.

Conclusions and Recomendations

Based on the analysis of fitbit data. Three recomendations can be made for the general marketing strategy for smartwatches.

1. Focus on the branding for activity tracking

Data shows that most users use the smartwatch for tracking their activity throughout the day rather than monitoring their sleep. It will be beneficial to capitalize on this and advertise the watch a watch able to do that. However, it can be noted that there is potential to also market the watch as a sleep tracker.

2. Emphasize ease of use and a good health tracking tool.

Smartwatch users wearing the watch do not perfrom high intensity for the large majority of the time. This means that the watch will be used less as a workout aid, but rather as a general health tracker. If it’s used in this setting, then it will be best for bellabeat to emaphsize the simplicty of the interface rather than it’s capabilities for advaced computations for athlete level tracking.

3. Market towards asthetics

Since most users spend long times wearing the watch while inactive and in a more causual setting, it might be good to focus efforts on emphasizing the asthetics of the watch while also stating the comfort of wearing the watch.

back to top