Customer Segmentation (K-Means) | Analysis

Author: Philip Wong (philipkfw@gmail.com)
Kaggle Source: https://www.kaggle.com/datasets/vjchoudhary7/customer-segmentation-tutorial-in-python

Project Context:

This case study uses a mock dataset hosted on Kaggle - intended to educate the concepts of customer segmentation, specifically unsupervised learning (K-Means Clustering).

Executive Summary:

Our 'client' runs a supermarket and has gathered basic data of their customers through issuing membership cards. Our goal is to better understand these customers in a way that will be useful for our client's marketing team. Some of the key questions we'll be looking to answer are;

1. Who are our target customers with whom our marketing team should focus their efforts?
2. How can we achieve customer segmentation using machine learning algorithm?

[1] Import Libraries

# data manipulation libraries
import pandas as pd
import numpy as np

# data visualization libraries
import matplotlib.pyplot as plt
import seaborn as sns
import plotly.express as px # interactive plots

# machine learning libraries
from sklearn.cluster import KMeans

# other libraries
from datetime import date
import os
import warnings
warnings.filterwarnings('ignore') # to remove any warning messages

[2] Data Load & Exploration

# import data
df = pd.read_csv("input/Mall_Customers.csv")

# explore dataset
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 200 entries, 0 to 199
Data columns (total 5 columns):
 #   Column                  Non-Null Count  Dtype 
---  ------                  --------------  ----- 
 0   CustomerID              200 non-null    int64 
 1   Gender                  200 non-null    object
 2   Age                     200 non-null    int64 
 3   Annual Income (k$)      200 non-null    int64 
 4   Spending Score (1-100)  200 non-null    int64 
dtypes: int64(4), object(1)
memory usage: 7.9+ KB

# verify if there's any missing values in each column
df.isna().sum()

CustomerID                0
Gender                    0
Age                       0
Annual Income (k$)        0
Spending Score (1-100)    0
dtype: int64

[3] Feature Engineering

# categorize ages by logical groups
criteria = [
    df['Age'] <= 18, # teen
    df['Age'] <= 25, # young adult
    df['Age'] <= 55, # adult
    df['Age'] <= 70, # senior
]
values = ['1. Teenager', '2. Young Adult', '3. Adult', '4. Senior']
df['Age_Category'] = np.select(criteria, values, default = '5. Retired')
df.head()

	CustomerID	Gender	Age	Annual Income (k$)	Spending Score (1-100)	Age_Category
0	1	Male	19	15	39	2. Young Adult
1	2	Male	21	15	81	2. Young Adult
2	3	Female	20	16	6	2. Young Adult
3	4	Female	23	16	77	2. Young Adult
4	5	Female	31	17	40	3. Adult

# categorize ages by 10-year interval buckets
criteria = [
    df['Age'] <= 10, 
    df['Age'] <= 20, 
    df['Age'] <= 30,
    df['Age'] <= 40,
    df['Age'] <= 50,
    df['Age'] <= 60,
    df['Age'] <= 70
]
values = ['1. 0-10', '2. 10-20', '3. 20-30', '4. 30-40', '5. 40-50', '6. 50-60', '7. 60-70']
df['Age_Bucket'] = np.select(criteria, values, default = '8. +71')
df.head()

	CustomerID	Gender	Age	Annual Income (k$)	Spending Score (1-100)	Age_Category	Age_Bucket
0	1	Male	19	15	39	2. Young Adult	2. 10-20
1	2	Male	21	15	81	2. Young Adult	3. 20-30
2	3	Female	20	16	6	2. Young Adult	2. 10-20
3	4	Female	23	16	77	2. Young Adult	3. 20-30
4	5	Female	31	17	40	3. Adult	4. 30-40

# add 2 new fields -> gender categorized by 1 or 0
df['Male Gender'] = np.where(df['Gender'] == 'Male', 1, 0)
df['Female Gender'] = np.where(df['Gender'] == 'Female', 1, 0)
df.head()

	CustomerID	Gender	Age	Annual Income (k$)	Spending Score (1-100)	Age_Category	Age_Bucket	Male Gender	Female Gender
0	1	Male	19	15	39	2. Young Adult	2. 10-20	1	0
1	2	Male	21	15	81	2. Young Adult	3. 20-30	1	0
2	3	Female	20	16	6	2. Young Adult	2. 10-20	0	1
3	4	Female	23	16	77	2. Young Adult	3. 20-30	0	1
4	5	Female	31	17	40	3. Adult	4. 30-40	0	1

[4] Data Visualization

Insight #1: Referring to the distribution plots below, there's a high number of customers aged between 19 to 39 years old with annual salaries ranging from \$25K to $80K

# plot distribution of numerical variables (age, annual income, & spending score)
plt.figure(figsize=(14, 5))
n = 0
for x in ['Age' , 'Annual Income (k$)' , 'Spending Score (1-100)']:
    n += 1
    plt.subplot(1, 3, n)
    plt.subplots_adjust(hspace=0.5 ,wspace=0.5)
    sns.distplot(df[x], bins=12, color='#2494E8')
    plt.title('Distribution of {}'.format(x))
plt.show()

Insight #2: Referring to the two plots below;

• Approximately 67% of customers are adults and 61% of adults are females
• Highest age bucket for both genders are 30-40 years old

# plot gender count
gender_count_df = df.groupby(['Gender', 'Age_Category']).size().reset_index(name='Count')
fig = (px.bar(gender_count_df, 
              x="Gender", 
              y="Count", 
              color="Age_Category", 
              text='Count',
              title="Breakdown of Female & Male Customers",
              color_discrete_sequence=["orange", "#F257EF", "#2494E8", "grey"],
              width=950, 
              height=400))
fig.show()

# let's investigate the distribution of age by 10-year age buckets
gender_count_df = df.groupby(['Gender', 'Age_Bucket']).size().reset_index(name='Count')
fig = (px.bar(gender_count_df, 
              x="Age_Bucket", 
              y="Count", 
              color="Gender", 
              text='Count',
              title="Breakdown of Female & Male Customers by Age Buckets",
              color_discrete_sequence=["#2494E8", "grey"],
              barmode="group",
              width=950, 
              height=400))
fig.show()

Insight #3: Let's see if there's any correlation between a customer's annual income relative to their age

• There is no correlation between these two variables

# Any correlation between annual income & age?
fig = (px.scatter(df, 
                  x="Age", 
                  y="Annual Income (k$)",
                  title='Correlation between Annual Income ($k) & Age',
                  color="Gender",
                  width=950, 
                  height=400))
fig.update_layout(legend_title_text='Age Category', )
fig.show()

Insight #4: Referring to the scatter plot below;

• There is a positive correlation between annual income and spending score (1-100), specifically among adults
• Young adults tend to have higher spending scores regardless of low annual incomes
• Potentially 5 different clusters

# Any correlation between annual income & spending score?
fig = (px.scatter(df, 
                  x="Annual Income (k$)", 
                  y="Spending Score (1-100)",
                  title='Correlation between Annual Income ($k) & Spending Score (1-100) by Age Category',
                  color="Age_Category",
                  hover_data = ['Age_Bucket'],
                  width=950, 
                  height=400))
fig.update_layout(legend_title_text='Age Category', )
fig.show()

Insight #5: Let's verify whether our correlation trends vary by gender

• It looks like the correlation between annual income and spending scores remain the same regardless of gender

# female
fig = (px.scatter(df[df['Gender']=='Female'], 
                  x="Annual Income (k$)", 
                  y="Spending Score (1-100)",
                  title='Female - Correlation Trend',
                  color="Age_Category",
                  hover_data = ['Age_Bucket'],
                  width=950, 
                  height=400))
fig.update_layout(legend_title_text='Age Category', )
fig.show()

# male
fig = (px.scatter(df[df['Gender']=='Male'], 
                  x="Annual Income (k$)", 
                  y="Spending Score (1-100)",
                  title='Male - Correlation Trend',
                  color="Age_Category",
                  hover_data = ['Age_Bucket'],
                  width=950, 
                  height=400))
fig.update_layout(legend_title_text='Age Category', )
fig.show()

# Correlation among numerical variables
numerical_columns = df.select_dtypes(include='int64').drop('CustomerID', axis=1)
plt.figure(figsize=(14, 5))
matrix = numerical_columns.corr()
sns.heatmap(matrix, 
            annot=True, 
            cmap="Blues")
plt.show()

[5] Modelling

Referring to the plot below - we'll be plotting the relationship between annual income and spending score in clustering into 5 distinct groups -> we will prioritize from 1 to 5 (1 == highest priority)

• [5] --- The cluster on the bottom left quadrant captures customers with lower income levels and lower spending habits - our marketing team should not prioritize this group
• [4] --- Meanwhile, the cluster on the bottom right quadrant captures customers who has high income levels, however understands the value of saving money without justifying any purchases that's classified as 'wants' - thus our client's marketing team should not prioritize this group
• [3] --- The cluster on the top left quadrant captures customers with low income levels, however enjoys shopping
• [2] --- The cluster located in the middle captures customers with moderate income levels and spending habits
• [1] --- The cluster on the top right quadrant clearly captures customers who has very high income levels which justifies their high purchasing habits - thus should be heavily prioritized by our client's marketing team

km = KMeans(n_clusters=5)
y_predicted = km.fit_predict(df[['Annual Income (k$)', 'Spending Score (1-100)']])
df['Cluster'] = y_predicted

# separate dataframes for each cluster
df1 = df[df.Cluster==0]
df2 = df[df.Cluster==1]
df3 = df[df.Cluster==2]
df4 = df[df.Cluster==3]
df5 = df[df.Cluster==4]

# plot scatter plot
["orange", "#F257EF", "#2494E8", "grey"],

plt.figure(figsize=(14, 5))
plt.scatter(df1['Annual Income (k$)'], df1['Spending Score (1-100)'], color='green', alpha=0.7, label='Cluster #1')
plt.scatter(df2['Annual Income (k$)'], df2['Spending Score (1-100)'], color='#2494E8', alpha=0.7, label='Cluster #2')
plt.scatter(df3['Annual Income (k$)'], df3['Spending Score (1-100)'], color='black', alpha=0.7, label='Cluster #3')
plt.scatter(df4['Annual Income (k$)'], df4['Spending Score (1-100)'], color='#F257EF', alpha=0.7, label='Cluster #4')
plt.scatter(df5['Annual Income (k$)'], df5['Spending Score (1-100)'], color='orange', alpha=0.7, label='Cluster #5')
plt.scatter(km.cluster_centers_[:,0],km.cluster_centers_[:,1],color='red',marker='*',label='centroid')
plt.xlabel('Annual Income (k$)')
plt.ylabel('Spending Score (1-100)')
plt.legend()
plt.show()

[6] Conclusion

• Based on our analysis, we can conclude that there is minimal difference between male and female customers, thus gender-based clustering will not lead to significant business impact
• Although majority of our customers are aged 25-40 years old, clustering by age may not lead to significant impact as we've established earlier that annual income has zero correlation with customer's age
• With that said, our data advises our shortage of customers with high-level incomes and high spending scores - an ideal strategy will be targeting these high-income customers