About this project
Bellabeat, founded in 2013 by Urška Sršen and Sando Mur, is a high-tech company focused on health products for women. It is a successful small company, but they have the potential to become a larger player in the global (https://en.wikipedia.org/wiki/Smart_device) device market. Click (https://bellabeat.com/) for more information about the company. Sršen, leveraging her artistic background, developed beautifully designed technology that tracks activity, sleep, stress, and reproductive health. This data-driven approach empowers women to understand their health and habits. By 2016, Bellabeat expanded globally, launching multiple products and opening offices worldwide. The company's products are available through online retailers and their e-commerce website. Bellabeat combines traditional advertising methods, such as radio, billboards, print, and TV, with a strong focus on digital marketing. They invest in Google Search, maintain active social media presence on Facebook, Instagram, and Twitter, and run video ads on YouTube and display ads through Google Display Network to support key marketing campaigns.
As a data analyst, the business problem I needed to solve was to analyze Bellabeat's consumer data to uncover insights into how users are interacting with their smart devices. Sršen wanted me to focus on understanding the usage patterns related to activity and sleep tracking. By analyzing this data, I aimed to identify trends in device usage, including which features are used most often, the time of day or circumstances when the devices are most active, and how these behaviors vary across different customer demographics.
Additionally, I looked for any gaps in usage or underutilized features. From this analysis, I generated high-level recommendations that could inform Bellabeat's marketing strategy. Ultimately, the goal was to help Bellabeat drive growth by aligning their product offerings and marketing efforts more closely with consumer habit. To do this, I followed the six steps of the data analysis process: ask, prepare, process, analyze, share, and act, to break down how I analyzed the (https://www.kaggle.com/datasets/arashnic/fitbit) in order to gain some insights that could be beneficial to Bellabeat.
Tools
I have used Python with libraries like Pandas for data manipulation, Matplotlib and Seaborn for visualization, and Jupyter Notebooks for an interactive coding environment. I have also used R with libraries such as dplyr for data manipulation, ggplot2 for visualization, and tidyr for data cleaning and preprocessing.
Steps
Data Exploration: Load and inspect the dataset to understand its structure and contents.
Cleaning & Preprocessing: Handle missing values, remove duplicates, and standardize date-time formats.
Exploratory Data Analysis (EDA): Visualize trends in activity, sleep, and calorie expenditure.
Correlation Analysis: Investigate relationships between activity levels, sleep duration, and heart rate patterns.
Tasks
In Python
Imported Excel datasheets with millions of entries into Python using pandas, then utilized libraries like pandas and NumPy to clean the Fitbit Fitness Tracker datasets and prepare them for analysis.
Merged datasets, created subsets of the data, and converted each attribute to the appropriate data type before performing analysis to gain insights into consumer usage of non-Bellabeat smart devices.
Developed visuals using Matplotlib, Seaborn, Plotly, and Tableau to display the analysis results, providing meaningful conclusions and actionable insights for stakeholders’ implementation.
In R
Importing Excel datasheets with millions of entries into R, then using tidyr package to clean the datasets and prepare them for analysis
Joining datasets together, creating subsets of the data, and converting each attribute to the proper data type before performing analysis to find differences between members and casual customers
Developing visuals using maltiplot, seaborn, plotly, ggplot2 packages, power BI and Tableau to display the analysis results and provide meaningful conclusions and insights for stakeholders’ implementation
The key stakeholders in this project included the following:
Urška Sršen: Cofounder and Chief Creative Officer at Bellabeat.
Sando Mur: Cofounder and key member of the Bellabeat executive team.
Marketing analytics team at Bellabeat: A team of data analysts responsible for collecting, analyzing, and reporting data that helps guide Bellabeat’s marketing strategy.
Customers: Everyone who purchases their product or use Bellabeat’s services.
My work was to analyze smart device usage data to gain insights into how consumers were using Bellabeat smart devices. I needed to understand broader trends that could be applied to Bellabeat’s own product offerings. After identifying these trends, I was to select one Bellabeat product and apply the insights to it for my presentation.
The analysis was guided by the following questions:
i. What were the key trends in smart device usage? ii. How could these trends be applied to Bellabeat’s customer base? iii. How could these trends influence Bellabeat’s marketing strategy? iv. The final deliverables for the report included:
A clear summary of the business task i. A description of all data sources used in the analysis ii. Documentation of any data cleaning or manipulation performed iii. A summary of my analysis and findings iv. Supporting visualizations to illustrate key insights v. My top high-level content recommendations based on the analysis
I followed the Case Study Roadmap as a guide and completed the case study within a week.
The data used is a free to use (https://www.kaggle.com/datasets/arashnic/fitbit) made available through Mobius. It contains personal fitness tracker data from over thirty FitBit users who have given consent to use their data. There are 18 csv files in all, but from the datasets I found only a few datasets relevant for my analysis. I thus focused on dailyActivity_merged.csv, hourlyCalories_merged.csv, hourlySteps_merged.csv, daily_Calories_merged.csv, dailySteps_merged.csv, dailyIntensities_merged.csv and sleepDay_merged.csv datasets_.
To quickly review the data, I opened each file in excel and observed that the data was structured in both wide and long formats. I also noticed that the dailyActivity_merged dataset included several metrics that could provide valuable insights, such as the total number of steps taken by Fitbit users, active minutes, and calories burned. These metrics could allow us to explore potential correlations, particularly between calories burned and steps taken. Additionally, the hourly calories and hourly steps datasets contained information on activity by hour, which would help provide insights into how calories and steps are distributed throughout the day.
Next, I organized the files by creating a separate folder on my Capstone Project folder, as I planned to use Python and Jupyter Notebook to process the data.
The credibility of Fitbit datasets: This depended on many factors, including data collection methods, device accuracy, and the consistency of the data over time. Here’s a breakdown of the factors that contribute to the data's credibility:
Data Collection: Fitbit devices collect data directly from users, capturing real-time activity metrics, such as steps, heart rate, calories burned, and sleep patterns. This makes the data reliable for tracking personal health metrics on a day-to-day basis, but the accuracy depends on users wearing the device consistently and correctly.
Device Accuracy: Fitbit devices use sensors such as accelerometers, heart rate monitors, and GPS for data collection. While Fitbit’s devices are generally accurate, some studies have shown that their step counters and calorie estimations can have slight discrepancies compared to medical-grade equipment.
Consistency: Fitbit datasets tend to be consistent for long-term usage, assuming users are regularly wearing their devices. However, variations in how often users sync their data, the type of activities they engage in, or how they use the device could introduce some inconsistencies.
Sampling Bias: Fitbit data is often self-reported through users who voluntarily choose to use the device, which means the data may not be representative of the general population. Users may also vary in how accurately they input information about their health habits.
Data Integrity: Fitbit datasets are typically well-maintained, with proper data formatting and storage practices. However, errors may still occur in syncing data or processing it into datasets, which should be handled during the analysis phase.
Overall, Fitbit datasets are generally credible for personal health insights and large-scale analysis, but potential issues related to device accuracy, consistency, and sampling bias should be considered when drawing conclusions from the data.
The Fitbit data was processed to extract meaningful insights about users' activities, sleep patterns, and health metrics. The following steps were used to process the datasets that includes activity logs, step counts, calories burned, sleep data, and other variables.
Here’s a general outline of the steps involved in processing Fitbit data:
The first step is to load the Fitbit data from a CSV file or other data format (e.g., .xlsx, .json) using libraries such as pandas. The data could include activity and sleep data over several hours or days.
Handling Missing or Inconsistent Data: Missing or corrupted data entries are identified and removed or filled. For example, if a column contains "NaN" or missing values, they can be dropped or filled using appropriate techniques. Datetime Conversion: Many Fitbit datasets include date and time columns (e.g., ActivityDate, ActivityHour, SleepDay). These columns need to be converted to datetime objects for easy manipulation and analysis.
Fitbit data is often recorded across different files, such as hourly steps, hourly calories, and sleep data. These datasets need to be merged based on common columns like Id (user ID) or ActivityHour (timestamp). Merging allows the datasets to be combined into one comprehensive dataset for further analysis.
After cleaning and merging the data, it's time to perform data analysis to extract insights. Some common analyses on Fitbit data include: Step counts: Summarizing total steps per day or over a specific time period. Calories burned: Calculating the total calories burned based on activity levels. Active minutes: Calculating time spent in different activity levels (e.g., sedentary, light active, very active). Sleep patterns: Analyzing the amount of time spent in deep sleep, light sleep, or awake. Active hours: Analyzing the hours of the day when users are most or least active. From the analysis, the following observations were made.
Observations
1. There is a positive correlation between Calories and TotalActiveMinutes (0.95), TotalSteps (0.94) and TotalDistance (0.94).
2. Many user used more time (minutes) in Light Active Minutes (84.7%) followed by Very Active Minutes (9.3%) and Fairly Active Minutes (6.0%)
3. Many device users participated in light activities and mostly covered light active distance followed by very active distance, moderately active distance with sedentary active distance as the last.
4. The burned calories increased by the total active hours. This is quite reasonable. The burned calories increased by the total hourly active hours. This is quite logical.
5. The burned calories decreased with sedentary minutes.
6. The burned calories increased with the total steps made by the users.
7. The calories burned increased with the active distance irrespective of the type.
8. This plot shows that the most calories were burnt on Tuesday and that the least calories were burnt on Sunday which is understandable because the users seem to be Christians they had not a lot of time for practice. However, Tuesday is rather strange because people seem to burn more calories than other days of the week. We needed to investigate why the users burned more calories on Tuesday.
9. Least steps were taken on Sunday. This could explain the least Calories burnt was recorded on Sunday. This could be because the surveyed users could be Christians and spent most of their time at home or in Church praying. Similarly, most steps were taken on Tuesday and that explains why most calories were burned on that day.
The data also gives us a clue about the profile of the users in the survey. They are most likely working class individuals.
10. More hours were taken on Wednesday for sleep and least on Monday. Since more steps were taken on Tuesday and thus the amount of calories burned on that day, the users could have been tired and took more time sleeping on Wednesday.
11. Many users seem to have more time in bed on Wednesday and least on Monday.
12. Tuesday was the most active day of the week and Sunday the least active day of the week.
13. For all the days of the week, most users covered light distance followed by very active distance, moderately active distance and lastly sedentary active distance
14. Tuesday has the highest very active distance covered by the users. This explains why there are more active steps on Tuesday and thus supports why most calories were burned on that day.
15. Most Bellabeat device users consume more sedentary minutes and so less active. This is followed by lightly active minutes, very active minutes and lastly fairly active minutes.
16. Most Bellabeat device users were involved in less active minutes. It is obvious that the users spend more time sitting or lying down, than they do being active. This can also reveal something about their occupation or lifestyle. Mostly likely they belong to the working class that spends most of their time behind their desks suggesting they could as well be mostly involved in online jobs.
17. Here we can see that their day starts getting really busy from 6 am in the morning all the way to 10 pm in the evening. The least active hours of the day are between 11 pm to 5 am. These are probably the best hours to reach them with targeted ads.
Explanation for the Observations
1. The positive correlation between Calories and TotalActiveMinutes, TotalSteps, and TotalDistance makes sense that burning calories aligns closely with activity levels. More movement, whether steps or distance, naturally increases calorie expenditure.
2. More time spent in Light Active Minutes (84.7%) shows that the Bellabeat device users engage mostly in light activity, with less time in moderate or vigorous activity. This may indicate a sedentary lifestyle or lower fitness levels.
3. Users cover mostly light active distance shows that there is preference for light activity that could be due to lifestyle or lack of structured workout routines.
4. The Calories burned increasing with total active hours shows an intuitive relationship, reinforcing the importance of staying active throughout the day.
5. The decreasing calories burned with sedentary minutes are due to extended sedentary periods that reduce energy expenditure, which could contribute to weight gain or other health issues.
6. The increase of calories burned with total steps shows that walking is a simple yet effective way to burn calories, and even small increases in step counts can improve health.
7. The increase of calories burned with active distance emphasizes the benefit of consistent movement.
8. Most calories being burned more on Tuesday and least on Sunday pattern may reflect workweek dynamics, with users more active during weekdays and resting on Sundays.
9. Least steps and calories burned on Sunday shows Sunday as being a low activity day, potentially linked to religious practices or rest days, reducing calorie burned.
10. More sleep on Wednesday, least on Monday could be due to high Tuesday activity that may lead to longer recovery sleep on Wednesday, while the transition from weekend to workweek disrupts Monday sleep.
11. More time in bed on Wednesday and least on Monday peoperly aligns with the previous point — users may be catching up on rest midweek.
12. Tuesday being the most active day is curious and worth exploring, while low Sunday activity could be linked to rest practices.
13. Light distance is the most common across all days shows that users favor light activity, likely due to lifestyle constraints or lack of structured fitness habits.
14. Highest very active distance was on Tuesday and this could explain the calorie burn spike on Tuesday-users might intentionally work out more on this day.
15. Users having high sedentary minutes points to prolonged periods of inactivity, possibly due to desk jobs or digital lifestyles.
16. Users spending more time inactive than active suggests a predominantly sedentary lifestyle may affect long-term health outcomes.
17. The peak activity being from 6 AM to 10 PM and least between 11 PM and 5 AM shows that Bellabeat device users follow a typical daily routine, with night hours naturally being less active.
Once the data is cleaned, merged, and analyzed, the results were visualized using matplotlib or seaborn to identify trends and insights. Visualizations like bar charts, line graphs, pie charts and scatter plots were used to display the relationships between variables such as steps, calories, sleep minutes, time in bed and active minutes.
Finally, insights were drawn based on the analysis. Some possible insights could include: Active times of the day: Identifying the hours when users are most and least active, Sleep habits: Analyzing how sleep durations correlate with activity levels or calories burned. User trends: Observing differences in activity across days of the week (e.g., more active on weekdays vs weekends).
Based on the processed Fitbit data, recommendations were made: These include: Increase Physical Activity: Since most users were sedentary, ways to increase activity throughout the day were recommended. Using active hours were recommended to push advertisements or notifications at optimal times to engage users when they are active.
Conclusion
Fitbit devices are excellent tools for tracking various aspects of daily health and fitness. The Bellabeat users used smart devices to track their steps, calories, active minutes, and sleep. By analyzing user data, we have provided valuable insights into users' activity and fitness patterns, allowing them to optimize their wellness routines and make informed decisions to enhance their overall well-being.
Further Questions to Explore
Why is activity level high on Tuesday?
Why is activity level unexpectedly not high on Saturday?
How do activity levels affect sleep quality?
Are there seasonal patterns in user activity?
Can we cluster users based on their health habits to create personalized recommendations?