ourabh-
S
Good morning. Today I'll be introducing our Project on Customer Segmentation for Big
[Link] challenge is: Big Bazaar has a diverse customer base with different shopping
behaviors, income levels, and spending patterns. A one-size-fits-all marketing approach isn't
[Link] solution is to use data-driven customer segmentation to identify specific customer
groups and tailor marketing strategies for each [Link]'re using customer purchase history
data - including annual income, spending score, and visit frequency - to develop a
Python-based clustering [Link] loaded Big Bazaar's customer data from a CSV file into a
pandas [Link] dataset contains important variables such as:
D
● emographics: Customer age, gender
● Behavioral data: Annual income, spending score, and visit frequency
Initial Data Exploration (EDA)
Before jumping into clustering, we needed to understand our dataset thoroughly.
Checking for Missing Values-For example, if some customers' income data was missing, it could
skew our analysis.
aanvi-
S
Examining Data Types-For instance, 'Annual Income' and 'Spending Score' should be
numerical, while 'Gender' should be categorical.
Generating Summary Statistics-This gave us valuable insights such as:
T
● he average annual income of Big Bazaar customers
● The typical spending score range
● The distribution of customer ages
ith a solid understanding of our dataset from the EDA phase, we were ready to move
W
into data preprocessing and feature selection
Punya-Data preprocessing
oving on to Data Preprocessing, this was a critical step to ensure the quality and fairness of
M
our clustering model. Our raw customer data had features like 'Age', 'Annual Income', and
'Spending Score', all on vastly different scales. If we used them as-is, a feature with a larger
range, like income, would unfairly dominate the model. To solve this, we used a technique called
Standardization with Python's StandardScaler. This process centers the data around a mean of
zero and scales it to a standard deviation of one. Essentially, it puts all our features on a level
playing field, ensuring that the distance calculations in the K-Means algorithm are balanced and
no single variable biases the formation of our customer segments.
Sourabh-Dimensionality Reduction
fter standardization, we proceeded with Dimensionality Reduction. Our dataset had multiple
A
dimensions, which are difficult to visualize and can sometimes contain noise. We used Principal
Component Analysis (PCA) to turn our original features into two new features (called
omponents) that aren’t related to each other. These two components keep the most important
c
information from the data. By reducing everything to just two dimensions, it becomes easier for
the clustering algorithm to work and also lets us plot the data on a 2D graph to see the clusters
clearly. In short, this step keeps the main patterns of the data but makes it simpler and easier to
understand.
TEJAS-optimal number of clusters
fter that, the K-Means clustering algorithm was applied with different values of k, which
A
represents the number of [Link] identify the best value of k, two methods were used — the
Elbow Method and the Silhouette [Link] the Elbow plot, the curve starts to bend around k
= 4 or 5, while the Silhouette Score is highest at k = [Link] both results, it can be
concluded that the optimal number of customer segments for Big Bazaar is likely between 2 and
4 clusters, giving a good balance between compactness and separation.
SAANVI-k-means clustering
-Means Clustering was the core technique used to perform Customer Segmentation Analysis. .
K
Feature Selection and Scaling (Data Preprocessing)
3. Applying K-Means Clustering
With k=4 determined, the actual clustering was performed:
Initialization: The K-Means algorithm was initialized with k=4 clusters.
itting the Model: The model was fitted to the scaled data. The algorithm iteratively assigns
F
each data point (customer) to the closest cluster center (centroid) and then recalculates the
centroid based on the new assignments. This continues until the centroids no longer move
significantly.
esult: Each customer in the dataset was successfully assigned to one of the four distinct
R
segments.
GUNMAY
fter performing clustering using the K-Means algorithm, our main task was to evaluate and
A
understand what each cluster represented. We analyzed the mean values of important features
such as Age, Annual Income, and Spending Score for each cluster.
his helped us identify distinct customer profiles. For instance, one cluster represented
T
customers with high income but low spending, showing a more saving-oriented behavior, while
another cluster included young customers with high spending, reflecting more outgoing or
impulsive habits.
y interpreting the clusters this way, we converted numerical data into meaningful customer
B
segments. These insights can help businesses create targeted marketing strategies, improving
engagement and revenue. for evaluation and interpreting
Punya-visualize clusters
ow, let's look at the results of our clustering in the Visualizations. This scatter plot is the direct
N
output of applying the K-Means algorithm to our preprocessed data. Each point on this plot
represents a customer, positioned based on the two principal components we created with PCA.
The colors you see represent the four distinct customer segments our model identified. A plot
like this is powerful because it allows us to see the natural groupings in our customer base. We
can immediately observe clear, well-separated clusters, which gives us confidence that our
model has found meaningful patterns and that the segments are distinct from one another.
Gunmay -Download results
nce the clusters were formed and interpreted, we added a new ‘Cluster’ column to our original
O
dataset, labeling each customer according to their segment. This made it easier to identify
which customer belonged to which group.
fter that, we exported the updated dataset into a CSV file using Python. This final step was
A
very important because it allowed us to save and share our analysis results for future use.
hese results can now be integrated with marketing tools or business applications to design
T
targeted campaigns for each customer group.
In short, downloading the results helped us preserve our analysis, making it accessible for
further decision-making and real-world implementation for download result
Tejas-Conclusion
hrough data preprocessing, PCA, and K-Means clustering, we effectively segmented Big
T
Bazaar’s customers into meaningful groups. The analysis provided valuable insights into
customer behavior and spending trends, enabling data-driven marketing strategies and
improved business decision-making.