Case Study: Analyzing Bumble Profiles¶

This notebook addresses various data analysis tasks including data cleaning, processing, and insights generation for Bumble profiles. The tasks are structured into four parts:

  1. Data Cleaning
  2. Data Processing
  3. Data Analysis
  4. Data Visualization

Part 1: Data Cleaning¶

This section focuses on inspecting missing data, fixing data types, and handling outliers. We'll also visualize missing data patterns to guide cleaning decisions.

In [2]:
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt

# Load the dataset
data = pd.read_csv('bumble.csv')

# Display basic information about the dataset
data.info()
# Show the first few rows of the dataset
data.head()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 59946 entries, 0 to 59945
Data columns (total 17 columns):
 #   Column       Non-Null Count  Dtype  
---  ------       --------------  -----  
 0   age          59946 non-null  int64  
 1   status       59946 non-null  object 
 2   gender       59946 non-null  object 
 3   body_type    54650 non-null  object 
 4   diet         35551 non-null  object 
 5   drinks       56961 non-null  object 
 6   education    53318 non-null  object 
 7   ethnicity    54266 non-null  object 
 8   height       59943 non-null  float64
 9   income       59946 non-null  int64  
 10  job          51748 non-null  object 
 11  last_online  59946 non-null  object 
 12  location     59946 non-null  object 
 13  pets         40025 non-null  object 
 14  religion     39720 non-null  object 
 15  sign         48890 non-null  object 
 16  speaks       59896 non-null  object 
dtypes: float64(1), int64(2), object(14)
memory usage: 7.8+ MB
Out[2]:
age status gender body_type diet drinks education ethnicity height income job last_online location pets religion sign speaks
0 22 single m a little extra strictly anything socially working on college/university asian, white 75.0 -1 transportation 2012-06-28-20-30 south san francisco, california likes dogs and likes cats agnosticism and very serious about it gemini english
1 35 single m average mostly other often working on space camp white 70.0 80000 hospitality / travel 2012-06-29-21-41 oakland, california likes dogs and likes cats agnosticism but not too serious about it cancer english (fluently), spanish (poorly), french (...
2 38 available m thin anything socially graduated from masters program NaN 68.0 -1 NaN 2012-06-27-09-10 san francisco, california has cats NaN pisces but it doesn&rsquo;t matter english, french, c++
3 23 single m thin vegetarian socially working on college/university white 71.0 20000 student 2012-06-28-14-22 berkeley, california likes cats NaN pisces english, german (poorly)
4 29 single m athletic NaN socially graduated from college/university asian, black, other 66.0 -1 artistic / musical / writer 2012-06-27-21-26 san francisco, california likes dogs and likes cats NaN aquarius english
In [3]:
data['last_online']
Out[3]:
0        2012-06-28-20-30
1        2012-06-29-21-41
2        2012-06-27-09-10
3        2012-06-28-14-22
4        2012-06-27-21-26
               ...       
59941    2012-06-12-21-47
59942    2012-06-29-11-01
59943    2012-06-27-23-37
59944    2012-06-23-13-01
59945    2012-06-29-00-42
Name: last_online, Length: 59946, dtype: object
In [4]:
data['last_online'] = pd.to_datetime(data['last_online'], format='%Y-%m-%d-%H-%M')
In [5]:
print(data['last_online'].isnull().sum())

0

Inspecting Missing Data¶

Check which columns have missing values and calculate the percentage of missing data in each column.

In [6]:
# Calculate missing values percentage
missing_data = data.isnull().mean() * 100
missing_data = missing_data[missing_data > 0].sort_values(ascending=False)
missing_data
Out[6]:
diet         40.694959
religion     33.740366
pets         33.231575
sign         18.443266
job          13.675641
education    11.056618
ethnicity     9.475194
body_type     8.834618
drinks        4.979482
speaks        0.083408
height        0.005005
dtype: float64

Dropping Columns with >50% Missing Data¶

In [7]:
# Drop columns where more than 50% of data is missing
data = data.loc[:, data.isnull().mean() <= 0.5]
data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 59946 entries, 0 to 59945
Data columns (total 17 columns):
 #   Column       Non-Null Count  Dtype         
---  ------       --------------  -----         
 0   age          59946 non-null  int64         
 1   status       59946 non-null  object        
 2   gender       59946 non-null  object        
 3   body_type    54650 non-null  object        
 4   diet         35551 non-null  object        
 5   drinks       56961 non-null  object        
 6   education    53318 non-null  object        
 7   ethnicity    54266 non-null  object        
 8   height       59943 non-null  float64       
 9   income       59946 non-null  int64         
 10  job          51748 non-null  object        
 11  last_online  59946 non-null  datetime64[ns]
 12  location     59946 non-null  object        
 13  pets         40025 non-null  object        
 14  religion     39720 non-null  object        
 15  sign         48890 non-null  object        
 16  speaks       59896 non-null  object        
dtypes: datetime64[ns](1), float64(1), int64(2), object(13)
memory usage: 7.8+ MB

Handling Missing Numerical Data¶

Impute missing values for numerical columns using the median of relevant categories.

In [8]:
# Define a function to impute missing numerical data by median within relevant groups
def impute_by_group(df, col, group_cols):
    df[col] = df.groupby(group_cols)[col].transform(
        lambda x: x.fillna(x.median())
    )
    return df

# Impute missing height and income by gender and status
data = impute_by_group(data, 'height', ['gender', 'status'])
data = impute_by_group(data, 'income', ['gender', 'status'])

Handling Outliers¶

In [7]:
# Replace -1 in income with 0
data['income'] = data['income'].replace(-1, 0)

# Calculate trimmed mean and median by removing extreme 10% of values
def trimmed_mean_median(series):
    lower, upper = np.percentile(series.dropna(), [10, 90])
    trimmed = series[(series >= lower) & (series <= upper)]
    return trimmed.mean(), trimmed.median()

# Example for height
mean_height, median_height = trimmed_mean_median(data['height'])
mean_height, median_height
Out[7]:
(68.23091633466136, 68.0)

Visualizing Missing Data¶

In [8]:
sns.heatmap(data.isnull(), cbar=False, cmap='viridis')
plt.title('Missing Data Heatmap')
plt.show()
No description has been provided for this image

Part 2: Data Processing¶

This section involves grouping continuous variables, creating derived features, and standardizing units for consistency.

Binning and Grouping¶

We will bin the age column into age groups and categorize income levels into meaningful thresholds.

In [9]:
# Binning age into age groups
age_bins = [18, 25, 35, 45, 100]
age_labels = ['18-25', '26-35', '36-45', '46+']
data['age_group'] = pd.cut(data['age'], bins=age_bins, labels=age_labels, right=False)

# Grouping income into categories
income_bins = [-1, 30000, 70000, 200000]
income_labels = ['Low Income', 'Medium Income', 'High Income']
data['income_group'] = pd.cut(data['income'], bins=income_bins, labels=income_labels, right=False)
data[['age_group', 'income_group']].head()
Out[9]:
age_group income_group
0 18-25 Low Income
1 36-45 High Income
2 36-45 Low Income
3 18-25 Low Income
4 26-35 Low Income

Derived Features¶

Create new features such as profile_completeness and convert height to cm.

In [10]:
# Calculate profile completeness
data['profile_completeness'] = data.notnull().mean(axis=1) * 100

# Convert height to cm
data['height_cm'] = data['height'] * 2.54
data[['profile_completeness', 'height_cm']].head()
Out[10]:
profile_completeness height_cm
0 100.000000 190.50
1 100.000000 177.80
2 84.210526 172.72
3 94.736842 180.34
4 89.473684 167.64

Part 3: Data Analysis¶

This section focuses on exploring demographic distributions, correlations, and patterns in the data.

Demographic Analysis¶

In [11]:
# Gender distribution
gender_counts = data['gender'].value_counts(normalize=True) * 100
gender_counts.plot(kind='bar', title='Gender Distribution', ylabel='Percentage')
plt.show()

# Status proportions
status_counts = data['status'].value_counts(normalize=True) * 100
status_counts.plot(kind='pie', autopct='%1.1f%%', title='Relationship Status Distribution')
plt.ylabel('')
plt.show()
No description has been provided for this image
No description has been provided for this image

Correlation Analysis¶

In [12]:
# Correlation between numerical columns
correlations = data[['age', 'income', 'height']].corr()
sns.heatmap(correlations, annot=True, cmap='coolwarm')
plt.title('Correlation Matrix')
plt.show()

# Age vs Income
sns.scatterplot(x='age', y='income', data=data)
plt.title('Income vs Age')
plt.show()
No description has been provided for this image
No description has been provided for this image

Diet and Lifestyle Analysis¶

In [13]:
# Dietary preferences
diet_counts = data['diet'].value_counts(normalize=True) * 100
diet_counts.plot(kind='bar', title='Dietary Preferences', ylabel='Percentage')
plt.show()

# Drinks vs Diet
sns.countplot(x='diet', hue='drinks', data=data)
plt.title('Drinks by Diet')
plt.show()
No description has been provided for this image
No description has been provided for this image

Geographical Insights¶

In [14]:
# Extract city and state from location
data['city'] = data['location'].str.split(',').str[0]
data['state'] = data['location'].str.split(',').str[1]

# Top 5 cities
top_cities = data['city'].value_counts().head(5)
top_cities.plot(kind='bar', title='Top 5 Cities with Most Users', ylabel='Number of Users')
plt.show()
No description has been provided for this image

Part 4: Data Visualization¶

Visualizing patterns to extract meaningful insights.

Age Distribution¶

In [15]:
# Plot age distribution
data['age'].plot(kind='hist', bins=10, title='Age Distribution', alpha=0.7)
plt.axvline(data['age'].mean(), color='red', linestyle='dashed', linewidth=1)
plt.xlabel('Age')
plt.ylabel('Frequency')
plt.show()
No description has been provided for this image

Income Analysis¶

In [16]:
# Boxplot of income by age group
sns.boxplot(x='age_group', y='income', data=data)
plt.title('Income by Age Group')
plt.show()
No description has been provided for this image

Analyzing last_online as a Date Field¶

Converting last_online into datetime format allows us to analyze user activity trends. We can derive:

  • The distribution of last active dates.
  • Most active days of the week.
  • Recency of user activity (e.g., within a week, month, etc.).
In [18]:
# Ensure 'last_online' is properly converted to datetime format
data['last_online'] = pd.to_datetime(data['last_online'], errors='coerce')

# Validate the conversion
print(data['last_online'].dtype)

# Proceed with datetime operations
data['last_online_date'] = data['last_online'].dt.date
last_online_distribution = data['last_online_date'].value_counts().sort_index()

# Plot distribution of last online dates
last_online_distribution.plot(kind='line', title='Distribution of Last Online Dates')
plt.xlabel('Date')
plt.ylabel('Number of Users')
plt.show()

datetime64[ns]
No description has been provided for this image

📊 Insights and Recommendations for Bumble Profile Analysis¶

1. Data Quality and Cleaning¶

  • Insight: Columns such as diet, religion, and pets have significant missing data (>30%). Additionally, income had placeholder values (-1) that needed replacement.
  • Recommendation:
    • Continue monitoring data collection for fields with high missing values.
    • Encourage users to complete these fields during profile setup or provide incentives for completion.

2. Demographic Insights¶

  • Insight: The majority of users fall within the age groups of 26-35 and 36-45. Gender and relationship status distributions show distinct patterns.
  • Recommendation:
    • Tailor marketing campaigns to these age groups for maximum engagement.
    • Create features that align with preferences observed in these dominant demographics.

3. Income and Age Correlation¶

  • Insight: There is a moderate positive correlation between age and income.
  • Recommendation:
    • Consider age-specific premium offerings tailored to financial capability.
    • Highlight user financial stability for transparency in profiles.

4. Dietary and Lifestyle Preferences¶

  • Insight: Social drinkers and users with a mostly anything diet dominate the data.
  • Recommendation:
    • Promote events or communities focusing on shared lifestyle preferences.
    • Refine diet-related filters in search functionalities.

5. Geographical Trends¶

  • Insight: User concentration is highest in major cities like San Francisco and Oakland.
  • Recommendation:
    • Regional campaigns should prioritize these high-user-density areas.
    • Explore features that facilitate local connections more effectively.

6. Profile Completeness¶

  • Insight: Profile completeness varies significantly, with many users missing key details.
  • Recommendation:
    • Introduce profile completion badges or incentives.
    • Provide prompts for completing critical profile sections.

7. Data Visualization Effectiveness¶

  • Insight: Visualizations provided clarity on missing data, demographic trends, and correlations.
  • Recommendation:
    • Integrate visual dashboards into operational analytics for continuous monitoring.
    • Use similar visual techniques for reporting to stakeholders.

Next Steps:¶

  1. Conduct A/B testing on incentives for profile completion.
  2. Explore machine learning models to predict user engagement based on demographics.
  3. Implement targeted ad campaigns for high-value segments.

By implementing these recommendations, Bumble can enhance user experience, improve data quality, and drive higher engagement across the platform. 🚀

In [ ]: