K-Mean Algorithm: Credit Card Customer Segmentation

In this guided project, we’ll play the role of a data scientist working for a credit card company. We’ve been given a dataset containing information about the company’s clients and asked to help segment them into different groups in order to apply different business strategies for each type of customer.

The company expects to receive a group for each client and also an explanation of the characteristics of each group and what are the main points that make them different.

In a planning meeting with the Data Science coordinator, it was decided that we should use the K-means algorithm to segment the data.

In order to use the algorithm properly and achieve all the goals that the company has set for us, we'll go through the following steps:

• Analyze the dataset;
• Prepare the data for modeling;
• Find an appropriate number of clusters;
• Segment the data;
• Interpret and explain the results.

We'll start by importing the packages we'll use.

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans

np.random.seed(42)

sns.set_style('whitegrid')
%matplotlib inline

Exploratory Data Analysis

After reading the data into pandas, it's time to explore it. Let's investigate the size of the dataset, what columns it contains, the type of values in each column, and also check on missing values.

customers = pd.read_csv('customer_segmentation.csv')

customers.head()

	customer_id	age	gender	dependent_count	education_level	marital_status	estimated_income	months_on_book	total_relationship_count	months_inactive_12_mon	credit_limit	total_trans_amount	total_trans_count	avg_utilization_ratio
0	768805383	45	M	3	High School	Married	69000	39	5	1	12691.0	1144	42	0.061
1	818770008	49	F	5	Graduate	Single	24000	44	6	1	8256.0	1291	33	0.105
2	713982108	51	M	3	Graduate	Married	93000	36	4	1	3418.0	1887	20	0.000
3	769911858	40	F	4	High School	Unknown	37000	34	3	4	3313.0	1171	20	0.760
4	709106358	40	M	3	Uneducated	Married	65000	21	5	1	4716.0	816	28	0.000

customers.shape

(10127, 14)

customers.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 10127 entries, 0 to 10126
Data columns (total 14 columns):
 #   Column                    Non-Null Count  Dtype  
---  ------                    --------------  -----  
 0   customer_id               10127 non-null  int64  
 1   age                       10127 non-null  int64  
 2   gender                    10127 non-null  object 
 3   dependent_count           10127 non-null  int64  
 4   education_level           10127 non-null  object 
 5   marital_status            10127 non-null  object 
 6   estimated_income          10127 non-null  int64  
 7   months_on_book            10127 non-null  int64  
 8   total_relationship_count  10127 non-null  int64  
 9   months_inactive_12_mon    10127 non-null  int64  
 10  credit_limit              10127 non-null  float64
 11  total_trans_amount        10127 non-null  int64  
 12  total_trans_count         10127 non-null  int64  
 13  avg_utilization_ratio     10127 non-null  float64
dtypes: float64(2), int64(9), object(3)
memory usage: 1.1+ MB

There are 10127 rows and 14 columns in the dataset including a unique identifier for each client, which is not going to be needed for the segmentation.

Of the 13 columns left, there are 8 columns containing integers, 2 containing floats, and 3 columns containing strings, which means we have 3 categorical columns to deal with.

Also, there are no missing values.

for col in ['gender', 'education_level', 'marital_status']:
    print(col)
    print(customers[col].value_counts(), end='\n\n')

gender
F    5358
M    4769
Name: gender, dtype: int64

education_level
Graduate         3685
High School      2351
Uneducated       1755
College          1192
Post-Graduate     616
Doctorate         528
Name: education_level, dtype: int64

marital_status
Married     4687
Single      3943
Unknown      749
Divorced     748
Name: marital_status, dtype: int64

Here we can see how many unique categories are there in each categorical variable and how many datapoints per category.

As we're working with unsupervised machine learning, there isn't a target variable on which we can measure the impacts of the other variables.

But we can see the correlation between the numeric variables and their distributions.

fig, ax = plt.subplots(figsize=(12,8))
sns.heatmap(round(customers.drop('customer_id', axis=1).corr(numeric_only=True), 2), cmap='Blues', annot=True, ax=ax)

plt.tight_layout()
plt.show()

Most of the variables present weak correlations between each other, but there are some we can highlight:

• Age is strongly correlated with how long the person has been a customer (months_on_book);
• Credit limit is positively correlated with the estimated income and negatively correlated with the average utilization ratio;
• The total number of transactions (total_trans_count) is strongly correlated with the total amount transitioned (total_trans_amount).

fig, ax = plt.subplots(nrows = 5, ncols = 2, figsize=(8, 10))

#Removing the customer's id before plotting the distributions
customers.drop('customer_id', axis=1).hist(ax=ax)

plt.tight_layout()
plt.show()

Regarding distributions, we have a couple of them closer to a normal distribution, but most of them are skewed.

Feature Engineering

We're now dealing with the 3 categorical variables.

The gender column is the easiest one. We'll use a lambda function to replace the values with ones and zeros.

We'll also be able to transform the education_level column to numeric. We'll use the replace() method to perform this task. This method will assign a value to each level of education:

• Uneducated - 0
• High School - 1
• College - 2
• Graduate - 3
• Post-Graduate - 4
• Doctorate - 5

Unfortunately, we can't do the same for this marital_status column as there is no level of magnitude between "Single", "Married" or "Divorced", for example. We can't say that any of them is higher or better than the others. Therefore, we'll use one-hot-encoding to create dummy variables from this column and then drop the original variable.

customers_modif = customers.copy()
customers_modif['gender'] = customers['gender'].apply(lambda x: 1 if x == 'M' else 0)
customers_modif.head()

	customer_id	age	gender	dependent_count	education_level	marital_status	estimated_income	months_on_book	total_relationship_count	months_inactive_12_mon	credit_limit	total_trans_amount	total_trans_count	avg_utilization_ratio
0	768805383	45	1	3	High School	Married	69000	39	5	1	12691.0	1144	42	0.061
1	818770008	49	0	5	Graduate	Single	24000	44	6	1	8256.0	1291	33	0.105
2	713982108	51	1	3	Graduate	Married	93000	36	4	1	3418.0	1887	20	0.000
3	769911858	40	0	4	High School	Unknown	37000	34	3	4	3313.0	1171	20	0.760
4	709106358	40	1	3	Uneducated	Married	65000	21	5	1	4716.0	816	28	0.000

# customers_modif.replace(to_replace={'Uneducated': 0, 'High School': 1, 'College':2,
#                                     'Graduate': 3, 'Post-Graduate': 4, 'Doctorate':5}, inplace=True)
customers_modif['education_level']=customers_modif['education_level'].map({'Uneducated': 0, 'High School': 1, 'College':2,
                                        'Graduate': 3, 'Post-Graduate': 4, 'Doctorate':5})
customers_modif['education_level'].head()

0    1
1    3
2    3
3    1
4    0
Name: education_level, dtype: int64

dummies = pd.get_dummies(customers_modif[['marital_status']], drop_first=True)

customers_modif = pd.concat([customers_modif, dummies], axis=1)
customers_modif.drop(['marital_status'], axis=1, inplace=True)

print(customers_modif.shape)
customers_modif.head()

(10127, 16)

	customer_id	age	gender	dependent_count	education_level	estimated_income	months_on_book	total_relationship_count	months_inactive_12_mon	credit_limit	total_trans_amount	total_trans_count	avg_utilization_ratio	marital_status_Married	marital_status_Single	marital_status_Unknown
0	768805383	45	1	3	1	69000	39	5	1	12691.0	1144	42	0.061	1	0	0
1	818770008	49	0	5	3	24000	44	6	1	8256.0	1291	33	0.105	0	1	0
2	713982108	51	1	3	3	93000	36	4	1	3418.0	1887	20	0.000	1	0	0
3	769911858	40	0	4	1	37000	34	3	4	3313.0	1171	20	0.760	0	0	1
4	709106358	40	1	3	0	65000	21	5	1	4716.0	816	28	0.000	1	0	0

Scaling the Data

First, we need to standardize the dataset. We'll use scikit-learn's StandardScaler() for this task.

$z = \dfrac{(x - \mu)}{\sigma}$

X = customers_modif.drop('customer_id', axis=1)

# Standardize the variables using z-score

scaler = StandardScaler()
scaler.fit(X)

X_scaled = scaler.transform(X)
X_scaled[:5]

array([[-0.16540558,  1.05995565,  0.50336813, -0.75221102,  0.1758098 ,
         0.38462088,  0.76394261, -1.32713603,  0.4466219 , -0.95970657,
        -0.97389518, -0.77588223,  1.07733799, -0.79850685, -0.28260887],
       [ 0.33357038, -0.9434357 ,  2.04319867,  0.66278684, -0.96716585,
         1.01071482,  1.40730617, -1.32713603, -0.04136665, -0.91643261,
        -1.35734038, -0.61627565, -0.92821381,  1.2523374 , -0.28260887],
       [ 0.58305837,  1.05995565,  0.50336813,  0.66278684,  0.78539682,
         0.00896451,  0.12057905, -1.32713603, -0.5736978 , -0.74098169,
        -1.91120566, -0.99715499,  1.07733799, -0.79850685, -0.28260887],
       [-0.78912553, -0.9434357 ,  1.2732834 , -0.75221102, -0.63697289,
        -0.24147306, -0.52278451,  1.64147829, -0.58525108, -0.95175829,
        -1.91120566,  1.75968594, -0.92821381, -0.79850685,  3.53845931],
       [-0.78912553,  1.05995565,  0.50336813, -1.45970995,  0.07421197,
        -1.86931731,  0.76394261, -1.32713603, -0.43087725, -1.05626345,
        -1.57036549, -0.99715499,  1.07733799, -0.79850685, -0.28260887]])

Choosing K

It's time to decide on the number of clusters. We'll run the k-means algorithm considering a range from 1 to 10 possible Ks and store the results. Then, we'll plot the elbow curve that will help us find a final K.

X = pd.DataFrame(X_scaled)
inertias = []

for k in range(1, 11):
    model = KMeans(n_clusters=k,n_init=10)
    model.fit_predict(X)
    inertias.append(model.inertia_)
    
plt.figure(figsize=(12, 8))
plt.plot(range(1, 11), inertias, marker='o')
plt.xticks(ticks=range(1, 11), labels=range(1, 11))
plt.title('Inertia vs Number of Clusters')

plt.tight_layout()
plt.show()

It looks like the rate of decreasing of the inertia slows down between 5 and 7 clusters. We'll use 6 clusters to move on.

model = KMeans(n_clusters=6, n_init='auto')
y = model.fit_predict(X_scaled)

y

array([5, 2, 5, ..., 3, 4, 3])

Analyzing Results

Now, let's create a CLUSTER column in our original dataset so we can better understand the characteristics of each one.

customers['CLUSTER'] = y + 1
customers

	customer_id	age	gender	dependent_count	education_level	marital_status	estimated_income	months_on_book	total_relationship_count	months_inactive_12_mon	credit_limit	total_trans_amount	total_trans_count	avg_utilization_ratio	CLUSTER
0	768805383	45	M	3	High School	Married	69000	39	5	1	12691.0	1144	42	0.061	6
1	818770008	49	F	5	Graduate	Single	24000	44	6	1	8256.0	1291	33	0.105	3
2	713982108	51	M	3	Graduate	Married	93000	36	4	1	3418.0	1887	20	0.000	6
3	769911858	40	F	4	High School	Unknown	37000	34	3	4	3313.0	1171	20	0.760	5
4	709106358	40	M	3	Uneducated	Married	65000	21	5	1	4716.0	816	28	0.000	2
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
10122	772366833	50	M	2	Graduate	Single	51000	40	3	2	4003.0	15476	117	0.462	1
10123	710638233	41	M	2	Graduate	Divorced	40000	25	4	2	4277.0	8764	69	0.511	2
10124	716506083	44	F	1	High School	Married	33000	36	5	3	5409.0	10291	60	0.000	4
10125	717406983	30	M	2	Graduate	Unknown	47000	36	4	3	5281.0	8395	62	0.000	5
10126	714337233	43	F	2	Graduate	Married	36000	25	6	2	10388.0	10294	61	0.189	4

10127 rows × 15 columns

customers['CLUSTER'].value_counts()

4    3132
3    2034
6    1880
2    1414
1     937
5     730
Name: CLUSTER, dtype: int64

numeric_columns = customers.select_dtypes(include=np.number).drop(['customer_id', 'CLUSTER'], axis=1).columns

fig = plt.figure(figsize=(20, 20))
for i, column in enumerate(numeric_columns):
    df_plot = round(customers.groupby('CLUSTER')[column].mean(),2)
    ax = fig.add_subplot(5, 2, i+1)
    ax.bar(df_plot.index, df_plot, color=sns.color_palette('Set1'), alpha=0.6)
    ax.set_title(f'Average {column.title()} per Cluster', alpha=0.5)
    ax.xaxis.grid(False)

    # Add data labels to the bars
    for i in ax.containers:
        ax.bar_label(i, label_type='edge')
    
plt.tight_layout()
plt.show()

For those numerical variables with higher correlations we saw earlier, we can also use a scatter plot to visualize this correlation grouped by clusters and analyze how the clusters change between each area of the chart.

fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(2, 2, figsize=(16, 8))
sns.scatterplot(x='age', y='months_on_book', hue='CLUSTER', data=customers, palette='tab10', alpha=0.4, ax=ax1)
sns.scatterplot(x='estimated_income', y='credit_limit', hue='CLUSTER', data=customers, palette='tab10', alpha=0.4, ax=ax2, legend=False)
sns.scatterplot(x='credit_limit', y='avg_utilization_ratio', hue='CLUSTER', data=customers, palette='tab10', alpha=0.4, ax=ax3)
sns.scatterplot(x='total_trans_count', y='total_trans_amount', hue='CLUSTER', data=customers, palette='tab10', alpha=0.4, ax=ax4, legend=False)

plt.tight_layout()
plt.show()

We can draw some early conclusions considering only the numeric variables.

For instance, Cluster 1 has the highest amount of money transitioned, while Cluster 2 has the lowest credit limit and estimated income and the highest utilization rate. Cluster 4 has the highest credit limit. Older clients are grouped in Cluster 5.

For the categorical columns, we'll plot the percentual distribution of each variable in each cluster. This will allow us to verify if a particular cluster is mostly composed of men, or of married people only, for example.

cat_columns = customers.select_dtypes(include=['object'])

fig = plt.figure(figsize=(18, 6))
for i, col in enumerate(cat_columns):
    plot_df = pd.crosstab(index=customers['CLUSTER'], columns=customers[col], values=customers[col], aggfunc='size', normalize='index')
    ax = fig.add_subplot(1, 3, i+1)
    plot_df.plot.bar(stacked=True, ax=ax, alpha=0.6)
    ax.set_title(f'% {col.title()} per Cluster', alpha=0.5)

    ax.set_ylim(0, 1.4)
    ax.legend(frameon=False)
    ax.xaxis.grid(False)
    
    labels = [0, 0.2, 0.4, 0.6, 0.8, 1]
    ax.set_yticks(labels)
    ax.set_yticklabels(labels)

    # Rotate x-axis labels by 0 degrees
    ax.set_xticklabels(labels=set(customers['CLUSTER']),rotation=0)

plt.tight_layout()
plt.show()

Considering the categorical variables, we notice that the education level is well divided between clusters.

In other highlights, Cluster 2 is composed almost entirely of married people, while we don't know the marital status of anybody in Cluster 3. Cluster 4 is almost completely male and Cluster 6 is 100% made of single people.

Conclusion

As demanded by the company, we now have listed the most important characteristics of each cluster. We could also some suggestions and insights into each one of them.

In the end, we have the list of customers with a cluster assigned to each one.

Cluster 1

Characteristics: Mostly men; high credit limit; high amount transitioned; high number of transactions; low utilization rate.

Insight: People with high volume spent on the card, but do not use it on a daily basis. Could be incentivised to spend more.

Cluster 2

Characteristics: Mostly women; mostly married; low estimated income; low credit limit; low amount transitioned; high utilization rate.

Insight: Married people (majority women) with low income and limit but utilize too much of their credit with a few larger purchases

Cluster 3

Cluster 3: Gender well divided; low credit limit, high utilization rate; marital status 100% unknown; smaller cluster.

Insight: Men and women with low credit limits but do have high balances.

Cluster 4

Cluster 4: Mostly men, mostly single and married, high estimated income, high credit limit; low amount transitioned; low utilization rate.

Insight: People (majority men) with high income and credit limits, but don't use the card. Could be incentivized to use it.

Cluster 5

Cluster 5: Mostly married, high age, low dependent count, long time customers, low credit limit, low amount transitioned, high utilization rate.

Insight: Older people and long-time customers. Low credit limit and transactions, but use the card very often. Could receive benefits to spend more money.

Cluster 6

Cluster 6: Mostly women; 100% single people, low estimated income, low credit limit, low amount transitioned, high utilization rate.

Insight: Single (mostly women) people that use their card a lot but have low credit limits and income. Could be given a bit more credit limit.

# List of customers and clusters
customers[['customer_id', 'CLUSTER']]

	customer_id	CLUSTER
0	768805383	6
1	818770008	3
2	713982108	6
3	769911858	5
4	709106358	2
...	...	...
10122	772366833	1
10123	710638233	2
10124	716506083	4
10125	717406983	5
10126	714337233	4

10127 rows × 2 columns