Customer Segmentation
Customer Segmentation
I hereby declare that the work presented in this report entitled “Customer Segmentation”
in partial fulfillment of the requirements for the award of the degree of Bachelor of
Technology in Computer Science and Engineering submitted in the Department of
The matter embodied in the report has not been submitted for the award of any other
degree or diploma.
This is to certify that the above statement made by the candidate is true to the best of my
knowledge.
(Supervisor Signature)
Supervisor Name: Dr. Ravindara Bhatt Designation: Assistant Professor
Department Name: Department of Computer Science and Information Technology
Date
i
ACKNOWLEDGEMENT
We would like to thank and express our gratitude to our Project supervisor Dr. Ravindara
Bhatt for the opportunity that he provided us with this wonderful project “CUSTOMER
SEGMENTATION”. The outcome would not be possible without his guidance. This
project taught us many new things and helped us to strengthen our concepts of Machine
Learning. Next, we would like to express our special thanks to the Lab Assistant for
cordially contacting us and helping us in finishing this project within the specified time.
Lastly, we would like to thank our friends and parents for their help and support.
ii
TABLE OF CONTENT
Chapter 1- Introduction
1.1 Introduction 1
1.2 Problem Statement 8
1.3 Objectives 9
1.4 Methodology 9
Chapter 5- Conclusions
5.1 Conclusion 50
References 34
iii
LIST OF FIGURES
Figure 12 – Heatmap 20
Figure 12 – Correlation 21
Figure 14 - Clustering 24
Figure 16 – Working of ML 31
iv
Figure 12 – Unsupervised learning 34
Figure 12 – Dataset 36
Figure 12 – Revenue 51
v
LIST OF TABLES
Table 1 6
Table 2 49
vi
ABSTRACT
Compelling choices are compulsory for any organization to produce good revenue. Nowadays
contest is huge and all organizations are moving forward with their own different strategies. We
ought to utilize information and take an appropriate choice. Each individual is different from the
other and we don't know what he/she purchases or what their likes are. However, with the
assistance of the machine learning method, one can sort out the information and can find the target
group by applying a few algorithms to the dataset. Without this, It will be very troublesome and no
better techniques are accessible to find the gathering of people with comparable person and
interests in an enormous dataset. Here, The customer segmentation utilizing K-Means clustering
assists with gathering the information with the same ascribes which precisely helps to business the
best. We are going to use the elbow technique to track down the number of clusters and finally, we
visualize the data.
vii
CHAPTER - 1 INTRODUCTION
1.1 Introduction
1
Customer Segmentation is the subdivision of a market into distinct client teams that share
similar characteristics. Customer Segmentation is a strong means that spot unsatisfied
customer requirements. Victimization on top of knowledge firms will then exceed the
competition by developing unambiguously appealing products and services.
Demographic Information, like orientation, age, familial and conjugal status, pay, training,
and occupation.
Geographical Information, which contrasts relying upon the extent of the organization. For
confined organizations, this data could relate to explicit towns or regions. For bigger
organizations, it could mean a client's city, state, or even nation of home.
Throughout the pretty long term, the opposition among organizations actually is expanded and
the enormous verifiable information that for all intents and purposes is accessible specifically
has brought about the inescapable utilization of information mining methods in removing the
significant and vital data from the data set of the association in a for all intents and purposes
big way. Information mining generally is the cycle where strategies basically are applied to
extricate information designs to basically introduce it in a generally intelligible arrangement
that can for the most part be utilized for the reason for the choice for the most part help in a
particular major way. As indicated, Bunching strategies specifically consider information tuples
as items. They segment the information objects into gatherings or groups so that items inside
a bunch kind of are like one another and unlike articles in different groups in a generally major
way.
2
Customer Segmentation mostly is the course of division of the client base into a generally few
gatherings called client sections to pretty such an extent that every client fragment comprises
clients who mostly have fairly comparative qualities, showing how throughout the definitely
long term, the opposition among organizations particularly is expanded and the enormous
verifiable information that basically is accessible essentially has brought about the inescapable
utilization of information mining methods in removing the significant and vital data from the
data set of the association, actually contrary to popular belief. The division depends on the
similitude in various ways that really are pertinent to promoting like orientation, age, interests,
and incidental ways of managing money, which actually is fairly significant. The client division
generally has the significance as it incorporates, the capacity to basically alter the projects of
the market with the kind of goal that it literally is kind of appropriate to every one of the client
portion, support in business choice; ID of items related with every client portion and mostly
manage the interest and supply of that item; distinguishing and focusing on the pretty potential
client base, and foreseeing client surrender, giving headings in viewing as the arrangements,
definitely contrary to popular belief.
The push of this paper kind of is to really recognize client sections utilizing the information
mining approach, utilizing the dividing calculation called as K-means grouping calculation,
which mostly shows that customer Segmentation particularly is the course of division of the
client base into a actually few gatherings called client sections to definitely such an extent that
every client fragment comprises clients who particularly have really comparative qualities,
showing how throughout the really long term, the opposition among organizations actually is
expanded and the enormous verifiable information that actually is accessible essentially has
brought about the inescapable utilization of information mining methods in removing the
significant and vital data from the data set of the association in a particularly big way.
The elbow strategy decides the fairly ideal groups, which specifically shows that as indicated,
Bunching strategies particularly consider information tuples as items. They segment the
information objects into gatherings or groups so that items inside a bunch essentially are like
one another and unlike articles in different groups, which basically is quite significant.
3
Fig 2-Customer Segmentation
● Supervised Machine Learning – In this, the data is labeled and also the algo learns
from labeled coaching data. Samples of this methodology area unit Classification and
Regression.
4
Often it is easier to induce untagged knowledge as compared to labeled knowledge, and in
such cases, we will use unattended machine learning to figure out the info. Data that
desires categorization is classified with the assistance of unattended machine learning.
The promoting “persona” is by definition a personification of a client section, and it's not
uncommon for businesses to form many personas to match their completely different client
segments.
But for that to happen, a business desires a sturdy set of client segments on which to base
it. that leads the United States of America to a successive section, identifying the
distinction between client segmentation and market segmentation, in order that your
segmentation is as correct as attainable.
5
Install dependencies
Table 1
6
Allowing your sales organization to pursue higher percentage opportunities:
By investing less energy in less worthwhile open doors and to a greater degree toward your
best portions, your outreach group will actually want to increment its success rate, cover
more ground, and at last increment incomes.
7
1.2 Problem Statement
Customer segmentation is characterized as "the method involved with separating clients
into bunches in light of normal qualities so organizations can market to each gathering
successfully and fittingly." Utilizing the right ascribes to characterize the client fragment, it
permits organizations to recognize the right clients to focus on and significant offers. The
individuals who effectively characterize and keep up with client division can get an upper
hand from the execution by further developing client experience.
Nonetheless, there are potential traps that can decrease the viability of a client division
drive. This article will distinguish the traps and propose arrangements to work on the
possibilities of a useful client division project.
Data Quality
The underlying arrangement of client division for organizations can be an obstacle.
Distinguishing the requirement for client division is the initial move towards carrying out a
cycle that lines up with your general marketable strategy. At the point when organizations
don't have a powerful client division process, they could end up offering a similar
assistance level for all clients and all items without zeroing in on the high-level clients or
items that acquire the best edges. To boost efficiency and benefit, the client division assists
organizations with applying the 80/20 rule, rather than extending themselves far by
attempting to offer a similar support level for each client, whether or not they are a top
client or not.
Challenge: Perhaps the greatest issue with client division is information quality. Wrong
information in source frameworks will normally bring about unfortunate gatherings. For
instance, clients who are people ascribed to age, orientation, and conjugal status are every
now and again utilized. On the off chance that these characteristics are not kept up with
appropriately, the fragments will be mistaken and subsequently, the data will probably be
less valuable. In the event that the clients feel awkward with the nature of the information,
they are probably not going to utilize the portions. Information quality issues additionally
emerge from an absence of support and ordinary purging to guarantee precision.
8
Solution: There are a few cycles that can be executed to give further developed information
quality to client division upkeep. One of the significant parts of information quality is the
idea of appointing assets to oversee credits for clients. This asset, typically called
information stewards, is answerable for dealing with the setup of another client, ensuring
all basic credits are given before the client is set up and support starts.
1.3 Objective
1.4 Methodology
9
Figure 3- Spiral Model
10
Four Phases of the Spiral Model are:
1. Planning:
The stage wherein prerequisites are recorded and chances are surveyed. During this stage,
we examined the venture title with the undertaking chief. Necessities and dangers were
evaluated after a writing search on one existing review and one more on the current review.
2. Risk Analysis:
The phase in which risks and alternative solutions are identified. At the end of this phase,
a prototype is created. If there is a risk in this phase, another solution is suggested.
3. Engineering:
4. Evaluation:
In this stage, the client plays out a product assessment. This is done after the framework is
introduced and clients test whether the framework lives up to their assumptions and
prerequisites. In the event that a mistake happens, the client can report the issue through
the framework.
11
1.4.1 Data Preprocessing
The preparation of data by using certain techniques before using it for prediction is known
as DATA PREPROCESSING.
Requirement:
Datasets are not generally fit to be utilized for investigation and forecast. They contain
commotion which implies undesirable qualities or undesirable credits which are not helpful
for us, missing qualities that influence our last response and might be available in an
undesirable configuration. In this way, information Preprocessing is expected to make the
dataset reasonable for examination.
a. Suitable Dataset
The as a matter of some importance prerequisite for an ML algorithm is a dataset in light
of the fact that a ML model generally works with just and just information. This
information is generally gathered for a chosen downside in a relevant arrangement and is
perceived as a dataset.
The informational collection can have many organizations for every single reason. For
instance, on the off chance that we will make a ML model for business purposes, the
informational index will be not the same as the informational collection required for client
division. So every single dataset accessible is not the same as the other datasets. To utilize
the dataset accessible in our code, we normally put it in a CSV record. In any case, here
and there we may likewise utilize HTML or xlsx records.
12
CSV File:
A Comma Separated Values (CSV) file is a delimited text file that uses commas to
differentiate values. In this, each line present is considered a data record. And these record
always have one or more fields, which is separated by commas. The name of the file
format is taken from the use of a comma as a file separator in a file. CSV files store data in
tabular form(numeric) in plain text, where each row contains the same number of fields.
b. Importing Libraries
For Data Preprocessing, certain libraries are very important. They needed to be imported.
They are:
NumPy:
Pandas:
It is a library in Python Programming language which is utilized for the control and
examination of information. It contains information designs and tasks with the assistance
of which we can control the mathematical tables and time series.
13
Matplotlib:
c. Importing Dataset:
We have selected a dataset from Kaggle and have read it with the help of a CSV file.
o Deleting Row: In this, we will find out which fields do not have values. We will
delete that record from the dataset. This method is not considered an efficient method as it
may lead to the loss of information.
o Calculating Mean: In this, we will calculate the mean of the column or row which
have missing values and replace the missing value with the mean, This method is mostly
used for the attributes which contain numeric data.
14
Figure 6- Checking missing values
● Dummy Variables
● Label Encoder
15
f. Data Cleaning:
g. Dataset is ready:
Figure 8- Dataset
16
h. Exploratory Data Analysis:
EDA is important in Unsupervised learning as it makes to have more domain knowledge.
This implies that people at age 35-36 do the most shopping at this mall.
17
Figure 10- Target Gender
After dummy variables, we know that 0=female and 1=male so females have had more
shopping in this mall.
18
It shows the relationship between Annual Income and Spending Score. and as it seems
people with the highest income have shopped the same or even less than average people.
a. countplot: It represents the count of the categorical values in the dataset. It is a part
of the seaborn library.
b. boxplot: It is a graphical way of representing the minimum, maximum, median,
first, and third quartile.
19
1.4.3 Creating the model (K-Means):
20
Figure 14- Correlation
21
1.4.4 Choosing K-value
22
Figure 15- Choosing K-value
Hence, k=5 is a good choice because we can see a significantly drop in the curve.
1.4.5 Building model with K-value:
23
Figure 16- Clustering
24
Figure 17- Analyzing Clusters
As we can see k=5 is a good choice for clustering here from elbow method..
Cluster 2 is people aged less than 40 with very high annual incomes, them having a high
spending score makes sense. So, to keep this going on, these people could be given better
offers to attract them.
Clusters 2 & 4 are the best choice to attract them with offers to buy from the ma
25
CHAPTER-2 LITERATURE SURVEY
26
2.2 Python
2.2.1 Introduction
Python is modern programming and figured out language. It upholds a few programming
ideal models, moreover as organized (particularly procedural), object-situated, and
deliberate programming. due to its escalated ordinary library, it's generally pronounced due
to the "battery-included" language. Guido van Rossum started creating Python.
Python is something anyway problematic to find out and its sentence is arranged all
together that reducing the costs to keep up the program. It supports modules and packs that
help the nature of the program and the reusability of code.
Routinely, we will quite often see by far most of the PC code designers' most appropriate
decision is Python. the point of its unmistakable quality is the immediate aftereffects of the
extended strength that it gives. Since there's neither any collection step in this manner it
presently makes it appallingly. The investigating that comes in Python is unimaginably
straightforward. Python programs are direct. At the point when an interpreter finds a
screw-up, it raises partner exclusion which is great. Regardless of direction, the program
can't get the exception, the work the interpreter just will is that it prints a stack follow. the
preeminent invigorating half is that the program itself is written in Python, that as of
presently shows anyway weighty the language is in itself. On the other hand, the speediest
course for working a program is to include not a few print explanations to the inventory.
27
2.2.2 History of Python
28
Python 2.0 (released on Oct sixteen, 2000), has several necessary new options like a cycle
detection dustman for memory management and support for Unicode.
29
2.3 Machine Learning
2.3.1 Introduction
ML could be a developing innovation that licenses PCs to gain from authentic information
naturally. ML utilizes a development of calculations to make numerical models and make
forecasts upheld by authentic information and data. it's as of now utilized for assorted
undertakings like picture acknowledgment, voice acknowledgment, email separating,
Facebook programmed labeling and recommender frameworks.
30
2.3.2 Working
ML frameworks gain from authentic data, assemble adumbrative models, and anticipate
their results once new information is gotten. The precision of the expected result relies
upon {the amount|the number|the amount} of information because the huge amount of
information helps construct piles of robust|an improved} model that predicts the result
extra precisely.
Assume you have a fancy disadvantage that believes you should make a few forecasts.
Hence, rather than composing code, you basically feed the information to normal
calculations, and these calculations are units used by the machine to frame rationale as
indicated by the information and foresee the result. ML has changed our procedure of
thinking about issues. The accompanying graph delineates how ML calculations work.
31
2.3.3 Need of ML
ML is changing into important step by step, the explanation required is that it'll perform
assignments that units of estimation are excessively refined for people to straightforwardly
carry out. As a human, there unit of estimation a few limitations as an aftereffect of you
can not physically access a lot of information. that wants some framework, and here is ML,
which could be rearranged.
You can prepare ML calculations by giving a lot of data, analyzing the information,
building models, and naturally foreseeing the predefined yield. The presentation of ML
calculations relies upon how much data and is not entirely set in stone by the value work.
ML helps in setting aside time and cash.
32
2.3.4 Types of Machine Learning
The target of supervised learning is to plan the info information to the result information.
Directed learning is instructor-based and is equivalent to when understudies learn things
under the oversight of an educator. An illustration of supervised learning is spam
separating.
▪ Regression
▪ Classification
33
2.3.4.2 Unsupervised ML:
Unsupervised learning is a technique wherein a machine learns without an educator.
Preparing is given to the machine utilizing a marked, arranged, or unclassified dataset, and
accordingly, the algorithmic program ought to answer this information while not
administration. The objective of unaided learning is to reproduce the info record into
groups of articles with new choices or comparable examples.
Unsupervised learning doesn't give pre-decided results. Getting information from immense
measures of data is exceptionally valuable. This sort of learning is isolated into two
algorithms:
● Clustering
● Association
34
2.3.4.3 Reinforcement ML:
Reinforcement learning might be an input-based learning procedure during which learning
specialists are compensated for all remedial activities and rebuffed for completely off-base
activities. The specialist precisely learns with this input and further develops execution. In
support of learning, specialists move with and investigate the environmental elements. The
specialist will probably procure the premier prize focuses and further develop execution.
A robot canine that naturally learns arm development is an illustration of support learning.
35
2.4 Dataset
Fig 24-Dataset
36
2.5 Libraries
1. Seaborn
The main aim is to visualize the central part of data understanding and exploration in a lot
of enticing ways. this is often supported by the core of matplotlib it's a library and also
provides a record-oriented API. This library is integrated with Panda's knowledge
structures, thus you'll be able to simply switch between totally different visual
representations of a specific variable to better understand the dataset to be used for
analysis.
2. Numpy
NumPy implies Numeric Python, a Python bundle for figuring and interacting
multi-layered and one-layered cluster parts. Travis Oliphant made the NumPy bundle in
2005 also due to the utility of the past Numeric module in another Numarray module.
This is a Python expansion module composed principally in C. differed capacities adjust
rapid mathematical computations. NumPy gives a spread of elite execution information
structures that carry out complex exhibits and frameworks. These DS region units are
utilized for ideal estimations on arithmetic.
37
3. Pandas
Pandas is characterized in Python as an open-source library that has strong data control.
The name of this library comes from the term board data, which suggests financial science
made out of two-layered data. Utilized for data examination in Python and created by Wes
McKinney in 2008. data investigation needs a store of cycles like z, python, panda, etc. Be
that as it may, I like pandas because of their are quicker, simpler, and much more
informative than various apparatuses.
Pandas are made on the NumPy bundle. All in all, NumPy is required for the panda to
figure. Before pandas were presented, Python was ready for data anyway had confined
help for data investigation. that is any place pandas came in, expanding the potential for
data examination. regardless of the inventory of the data, you'll play out the 5 key advances
expected to technique and dissect the data. NS. Stacking, activity, arrangement, displaying,
and examination.
4. Sklearn
Scikit-learn also referenced as sklearn, was called scikit. Learning can be a free AI
programming framework library for the Python fake language. SVM, irregular woods,
inclination supporting, k means, DBSCAN, and quite a bit of other grouping, relapse, and
pack calculations are intended to work with the mathematical and logical Python libraries
NumPy and SciPy. Increment. Scikit-learn can be an undertaking financed by
NumFOCUS.
38
CHAPTER-3 SYSTEM DEVELOPMENT
39
3.2 Algorithms Used
Clustering algorithms generate clusters within the cluster area unit that similarly support
some characteristics. The similarity is outlined in terms of how close the object area unit in
the house is.
K-means algorithm program in one of all the foremost well-liked centers of mass-based
algorithmic program. Suppose the information set, D, contains n objects in the house.
Partitioning strategies distribute the objects in D into k clusters, C1,..., Ck, that is, Ci ⊂ D
and Ci ∩Cj = ∅ for (1 ≤ i, j ≤ k). A centroid-based partitioning technique uses the center of
mass of a cluster, Ci, to represent that cluster. Conceptually, the center of mass of a cluster
is its central purpose. The distinction between an object p ∈ Ci associated with ci, the
representative of the cluster, is measured by dist(p, ci), wherever dist(x,y) is the Euclidean
distance between 2 points x and y.
Algorithm: The k-means rule for partitioning, where every cluster’s center is portrayed
by the mean value of the objects within the cluster. Input: k: the number of clusters, D:
information set containing n objects. Output: a group of k clusters.
Method: (1) haphazardly select k objects from D because the initial cluster centers; (2)
repeat (3) (re)assign every object to the cluster to that the thing is that the most similar,
supported the mean of the objects within the cluster; (4) update the cluster suggests that
that is, calculate the mean of the objects for every cluster; (5) till no change.
40
Fig 26-5 clusters
The elbow method actually is predicated on the observation that increasing the number of
clusters will specifically facilitate cutting back the actual total within-cluster variance of
every cluster in a particularly major way. This basically is as a result of having additional
clusters particularly permit one to for all intents and purposes capture finer teams of an
information object that really are additionally similar to one another, really contrary to
popular belief. To outline the definitely optimum clusters, Firstly, we use the clustering
algorithmic program for numerous values of k. This literally is done by going k from one
to ten clusters, so the elbow method literally is predicated on the observation that
41
increasing the number of clusters will essentially facilitate cutting back the sort of total
within-cluster variance of every cluster in a fairly big way. Then we actually calculate the
basically overall intra-cluster very total of sq in a subtle way. Then, we kind of proceed to
plot the intra-cluster kind of total of sq, or so they essentially thought. based mostly on the
number of clusters, so based mostly on the number of clusters, which specifically is fairly
significant. The plot denotes the particularly approximate variety of clusters needed in our
model, so then, we specifically proceed to plot the intra-cluster for all intents and purposes
total of sq, or so they mostly thought. The basically optimum clusters will mostly be really
found from the graph where there's a bend within the graph in a definitely major way. The
kind of optimum clusters can kind of be kind of found from the graph where there actually
is a bend in the graph, showing how based mostly on the number of clusters, so based
mostly on the number of clusters in a fairly big way.
42
3.2.3 Customer Segmentation
Throughout the long term, as there explicitly is extremely strong contest inside the
business world, the associations certainly have been constrained to for the most part
improve their benefits and business by and large fulfilling the kind of their clients and
drawing in new clients in accordance with their requirements, which truly is very critical.
The distinguishing proof of clients and kind of fulfilling the type of every buyer can
generally be a for the most part progressed and monotonous assignment, which generally is
very huge. Generally, this can be frequently because of clients most certainly is
additionally most certainly entire totally unique to truly keep with their requests, tastes,
inclinations, etc, which especially is very critical. Rather than the "one-size-fits-all"
approach, buyer division bunches the customers into bunches sharing indistinguishable
properties or social qualities in an unobtrusive manner. to in every practical sense, keep
with, client division can in a real sense be a technique for separating the market into
homogeneous gatherings, exhibiting that throughout the long term, as there truly is
exceptionally strong rivalry inside the business world, the associations in every way that
really matters, have been constrained to sort of upgrade their benefits and business by
especially fulfilling the type of their clients and drawing in new clients in accordance with
their very in an unpretentious way. the information used inside the shopper division
method that separates the buyers into groups relies upon shifted factors like information on
topographical circumstances, financial circumstances, segment conditions moreover, and
movement designs, exhibiting how throughout the long term, there generally is extremely
durable contest inside the business world, the associations basically have been constrained
to improve their benefits and business by essentially fulfilling the type of their clients and
drawing in new clients in accordance with their necessities, or so they for the most part
thought. The purchaser division strategy for the most part allows the business to frame
certainly higher utilization of their advancing spending plans, gain an upper hand over their
fundamentally rival firms, and especially exhibit the fundamentally upper information on
the prerequisites of the shopper, which basically shows that throughout the long term, as
there in every practical sense, is extremely solid contest inside the business world, the
associations, for the most part, have been constrained to truly improve their benefits and
business by really fulfilling the type of their clients and drawing in new clients in
43
accordance with their sort of requirements in an unobtrusive manner. It basically set up
truly assists an association with expanding its advancing proficiency, decides new market
open doors, truly makes the extremely next fundamentally entire procedure, recognizing
clients maintenance, exhibiting that the buyer division method generally allows the
business to frame essentially higher utilization of their advancing spending plans, gain a
strategic advantage over their especially rival firms, and sort of show the truly upper
information on the prerequisites of the shopper, which in a real sense shows that
throughout the long term, as there really is exceptionally durable contest inside the
business world, the associations truly have been constrained to essentially upgrade their
benefits and business by exceptionally fulfilling the kind of their clients and drawing in
new clients in accordance with their very needs in an inconspicuous manner.
Assumptions:
1. Variables must have actual value otherwise the classifier will give a guessed result
in place of an accurate result.
2. There should be low correlations between the predicted results of trees.
44
Working of the algorithm:
45
CHAPTER-4 PERFORMANCE ANALYSIS
4.1 Comparison
The most common ways in which within which businesses phase their client base are:
● Demographic information, like gender, age, familial and legal status, income,
education, and occupation.
● Geographical information differs counting on the scope of the corporate. For
localized businesses, this information would possibly pertain to specific cities or counties.
For larger corporations, it'd mean a customer’s town, state, or perhaps the country of
residence.
● Psychographics, like socio-economic class, lifestyle, and temperament traits.
● Behavioral data, like outlay and consumption habits, product/service usage, and
desired edges.
46
47
Fig 29- Comparisons
48
Table 2
49
CHAPTER-5 CONCLUSIONS
5.1 Conclusions
● Cluster one denotes the clients with medium financial gain and medium payment
scores this sort of customers may or may not be beneficial.
● Cluster two denotes the low annual financial gain however high yearly payments
this sort of customers can be attracted by giving them minimal expense EMI's.
● Cluster three denotes a high annual financial gain moreover a high yearly payment
so, we can focus on these sorts of clients as they bring in more cash and spend how
much they need.
● Cluster four represents the cluster having high annual financial gain and low annual
pay so, we can focus on these sorts of clients by asking for feedback and promoting
the item in a superior manner.
● Cluster five represents customers with low annual financial gain and low annual
payment, there is no need to target this kind of customers as they earn less and
spend less
● From the customer map we can see that most of the customers are from India,
USA, Japan, and United Kingdom.
● From Revenue share diagram we can see that the highest revenue contributing
countries are India, USA, Japan, and United Kingdom.
50
Fig 30- Customer Map
51
Fig 32- Final Clusters
52
REFERENCES
● Sari, J. N., Nugroho, L. E., Ferdiana, R., & Santosa, P. I. (2016). Review on
Customer Segmentation Technique on Ecommerce. Advanced Science Letters, 22(10),
3018–3022. https://doi.org/10.1166/asl.2016.7985
● Wu, J., Shi, L., Lin, W. P., Tsai, S. B., Li, Y., Yang, L., & Xu, G. (2020). An
Empirical Study on Customer Segmentation by Purchase Behaviors Using a RFM Model
and K -Means Algorithm. Mathematical Problems in Engineering, 2020.
https://doi.org/10.1155/2020/8884227
53