Journal of Applied Data Sciences ISSN 2723-6471

Vol. 2, No. 1, January 2021, pp. 19-25 19

Maximizing Strategy Improvement in Mall Customer Segmentation

using K-means Clustering

Musthofa Galih Pradana 1,*, Hoang Thi Ha 2,

1
Department of Informatics Alma Ata University, Yogyakarta, Indonesia
2
Department Management Information System University of Danang, Vietnam
1
mgalihpradana@almaata.ac.id; 2 hoang2th@due.udn.vn
* corresponding author

(Received December 21, 2020 Revised January 4, 2021, Accepted January 14, 2021, Available online January 15, 2021)

Abstract
The application of customer segmentation is very vital in the world of marketing, a manager in determining a marketing
strategy, knowing the target customer is a must, otherwise it will potentially waste resources to pursue the wrong target.
Customer segmentation aims to create a relationship with the most profitable customers by designing the most
appropriate marketing strategy. Many statistical techniques have been applied to segment the market but very large data
are very influential in reducing their effectiveness. The aim of clustering is to optimize the experimental similarity within
the cluster and to maximize the dissimilarity in between clusters. In this study, we use K-means clustering as the basis for
the segmentation that will be carried out, and of course, there are additional models that will be used to support the
research results. As a result, we have succeeded in dividing the customer into 5 clusters based on the relationship
between annual income and their spending score, and it has been concluded that customers who have high-income levels
& have a high spending score are also very appropriate targets for implementing market strategies.
Keywords: Segmentation, Strategy, Clustering, K-Means.

1. Introduction
In this era, increasing the level of consumer consumption is very reasonable, this is based on the very fast
development of production. This makes each person feel like they have an obligation to spend something to
enjoy these developments. At this point, an increase in the number and variance of products is not a bad thing
for the market, but an increase in customers can sometimes lead to wasted resources due to a strategy aimed at
the wrong customer [1]. At this time a lot of managers and people who work in the marketing field try various
things to create the right market strategy. However, we are talking about their customers who are human and
change or can change based on various factors. Many applications of certain strategies such as discounts,
annual promotions, memberships, etc. may work for a while but after that, it is nothing more than a waste of
resources, both energy, and money.
As a manager, it is very important to be able to recognize the patterns or habits of the customers themselves.
As a matter of fact, the mall Industry is often involved in a race to increase their customers and therefore make
huge profits [2]. There are several factors that explain why the mall rejects its role. First, the level of customer
activity is higher, they have less time to shop, and finally, they reduce their shopping frequency. In fact, there
are too many of the same malls in a district or city and eventually, customers will go to the shopping center
that offers the most products and the best service. This factor encourages mall managers to develop a strategy
to differentiate them from competitors [3, 4].

2. Literature Review
research shows that detecting where a customer is going to meet their shopping needs is highly dependent on
the service from the provider and the characteristics of the place they are going to. In certain perspectives such

Musthofa Galih Pradana and Hoang Thi Ha / JADS Vol. 2 No. 1 2021
Journal of Applied Data Sciences ISSN 2723-6471
Vol. 2, No. 1, January 2021, pp. 19-25 20
as McGoldrick and Thompson they indicate that the level of price, crowd, convenience, and service are very
vital factors. grouping mall classes or categories into functional types, recreation areas, social places & public
places is very important. This is used by some mall managers / owners in determining target visitors and the
types of services they provide.
Using quite different perspectives, Lehew and Wesley categorize the mall section as multiple neighborhoods
and values. Anselmsson said that the most important factor in fulfilling customer satisfaction is the selection
of the right atmosphere, equipment, promos, and communicative methods. With this, it is hoped that this type
of grouping will provide the best service & experience for customers who come. Previously we mentioned
several types of grouping of important factors that malls should pay attention to, but customers must also be
able to be grouped. This is necessary because most methods or strategies will fail if the target customer group
is wrong. In this research, we will classify mall customers using machine learning methods in order to get a
clear visualization of the existing customer groupings.
2.1. Machine Learning
We have often seen the application of machine learning in various fields around us, for example, on Facebook,
machine learning helps us to identify ourselves and our friends, or even on YouTube recommending videos
based on the things you are interested in. Machine learning itself is generally categorized into 2 types, namely
Supervised and Unsupervised learning. Supervised learning is usually used by a data analyst to solve
problems such as classification and regression [5, 10], which means that in this case the data there is a target
label that you want to predict in the future, for example predicting the value of a student or the number of
monthly expenses. On the other hand, in unsupervised learning, its users do not always have a special label or
target to predict, for example, clustering, based on its mathematical model, the algorithm in unsupervised
learning does not have a target of a variable [6]. For example, we want to classify students based on their
learning habits or create clusters based on the number of purchases of a product.
The marketing industry, especially the malls, has tough competition to increase their customers and therefore
generate huge profits. To achieve this task, machine learning is already being implemented by many shops
and other markets [7]. malls or shopping centers take advantage of the data they get when transacting with
their customers and make use of it by developing ML models to target the right ones [8,9]. This not only
increases sales & the number of visitors who come but also increases efficiency in doing business.
2.2. Clustering
Clustering is known as a method for identifying common groups in a data set. The entities in each group are
comparatively more similar to entities from that group than to other group entities. Since the 70s, cluster-
based segmentation has been used very often in various studies involving data, especially in marketing. As
stated by [11], that clustering is not a structured method of data analysis, even though it has good flexibility, it
really depends on the data or sample used. A statistical approach used by some studies with cluster analysis by
[12], it is called the "tandem method" consisting of two processes the first is factor analysis and the second is
performing cluster analysis.
This approach has been heavily criticized by several other articles. It is all caused by a key problem in his
method, namely preliminary factor analysis can destroy existing cluster structures [13]. As an alternative to
the tandem method, hierarchical cluster analysis can be used as an alternative by using binary variables.
However, the reliability of this method was highly questioned by many researchers at its time and non-
hierarchical methods have been very dominant since the 80s.

Musthofa Galih Pradana and Hoang Thi Ha / JADS Vol. 2 No. 1 2021
Journal of Applied Data Sciences ISSN 2723-6471
Vol. 2, No. 1, January 2021, pp. 19-25 21

3. Method
Table. 1. Dataset Structure

CustomerID Gender Age Annual Income (k$) Spending

Score (1-100)

0 1 Male 19 15 39

1 2 Male 21 15 81

2 3 Female 20 16 6

3 4 Female 23 16 77

4 5 Female 31 17 40

First of all the research begins by knowing what kind of data we use, see table 1 for the dataset. The dataset
we use is quite simple but very detailed, consisting of customer ID, gender, age, annual income, and purchase
score. What is meant by a spending score is the value of how much the customer shopped or spent their
money at the mall, the value is on a scale from 1 to 100 (higher means the more is spent). The structure of the
dataset has been displayed properly, but what about the contents of the data are there any missing values? We
are lucky that there is no missing data in our dataset, see Figure 1 below for the results of the total values.

Figure. 1. Dataset Value

After knowing the data we have, we can do the plotting, based on Figure 2 below, we do this by comparing
the annual income and spending score, which is of course differentiated by gender. From the results we get we
see that there are customer behaviors with annual income and Spending scores, there are 5 type plot shows
segments of Customers with the following behaviors:
● Low Income- Low Spending Score
● High Income- Low Spending Score
● Low Income- High Spending Score
● High Income- High Spending Score
● Average Income- Average Spending Score

Musthofa Galih Pradana and Hoang Thi Ha / JADS Vol. 2 No. 1 2021
Journal of Applied Data Sciences ISSN 2723-6471
Vol. 2, No. 1, January 2021, pp. 19-25 22

Figure. 2. Annual Income vs Spending Score

Knowing that there are already several groupings, although not in detail we can make a K-means model now.
The method we will use is The Elbow method is a really common method as well as the concept would be to
perform k-means clustering for a range of k clusters (let's say 1 to 10) and to measure the sum of square
distances from each point to its assigned center for each value. The mean squared distance between each
instance and its nearest centroid is defined as this [17]. Analytically, the lower inertia of the model is higher,
as per the description. We observed that the point after this there is no sudden change in WCSS (Within
Cluster Sum of Squares) is found in K=5. And we're going to use K=5 as the right number of clusters that fit
the clusters we had before the algorithm was implemented. See figure 3.

Figure. 3. Elbow method result

By using the above method, we can divide the plot into several clusters and find out which clusters can be
prioritized, and give the appropriate label to each cluster. By using the K-means algorithm, we can find out
which of the five clusters should be targeted, namely customers with Moderate Income- Moderate Spending
Score, High Income- High Spending Score, Low Income- High Spending Score. As we can see in figure 4, the
targeted customers have been found.

Musthofa Galih Pradana and Hoang Thi Ha / JADS Vol. 2 No. 1 2021
Journal of Applied Data Sciences ISSN 2723-6471
Vol. 2, No. 1, January 2021, pp. 19-25 23

Figure. 4. Final cluster of clients

4. Results and Discussion

We can see that, based on their annual income and spending score, mall customers can be grouped into 5
groups. First, the green group indicates that they are people with high incomes and high spending scores, this
is an ideal target for a mall or shopping district because people like this are the biggest and most potent source
of profit. In fact, this person may be a regular visitor to a mall and easily be convicted by the mall facilities.
Second, the blue group we can know is a group of people who have a high income but have a low level of
spending. This is a very interesting case, given the many potential reasons for such a group to exist. for now,
let's assume that they are people who are very active in shopping but are not satisfied with the services or
facilities in the mall. such groups are also a good potential target, but we need to be able to identify in advance
the reasons for their low levels of spending. the department manager or mall authority can develop or add a
facility & offer that can help attract groups of visitors like this to come and have their needs met.
Third, the red group from the data we get identifies them as having average incomes and spending levels. we
can assume these are people who don't always buy a product but have a high level of willingness to spend
even though sometimes they have little income. This group of people is not a group that has a high potential
high income for the mall, and also as a manager as much as possible to avoid targeting this group of people in
a market strategy. however, they can still be considered through other data analysis techniques that might
increase their level of spending.
fourth, the cyan-colored group, as we can see, this group contains people who have low income but have high
spending scores, people like this have pleasure or hobby in spending something even though they have a low
income. This is also possible if they are people who feel comfortable or satisfied with the services provided by
the mall so that they feel compelled to spend something because the service makes them satisfied.
Fifth, the yellow group classifies people who have low annual incomes and low spending scores. and it is
quite reasonable also that they have low income so they will spend less on something, even what they do
maybe a wise and good choice based on their condition. a mall manager should target the people in this cluster
at the lowest priority.

Musthofa Galih Pradana and Hoang Thi Ha / JADS Vol. 2 No. 1 2021
Journal of Applied Data Sciences ISSN 2723-6471
Vol. 2, No. 1, January 2021, pp. 19-25 24
Based on their Annual Income and Spending Score, we know the behavior of customers by looking at the
results. Many marketing tactics for customers can be adapted to this cluster study. Our target customers are
high income and high spending score customers, and we will still like to keep them as they offer the most
profit margin. Customers with a wide range of products will be attracted to their lifestyle requirements for
high income and lower spending score and that could attract them to the Mall Supermarket. Less Income Less
Spending Score can be sent additional promotions and they will be drawn to spending by continuously giving
them offers and discounts. A cluster analysis can also be performed on what kind of products consumers
choose to consume and can find other marketing campaigns accordingly.

5. Conclusion
This research proves that it is possible to do segmentation on customers in malls. even the application of
machine learning like this is very profitable in the industry, a manager can pay full attention to handling each
cluster that has been identified by meeting their every need. To meet the needs of customers, mall managers
must be able to understand what is needed and be in the minds of customers, study their shopping habits and
maintain regular interactions with customers that can make them feel comfortable.
This research proves that it is possible to implement machine learning in the industrial segmentation of this
shopping district. But assuming machine learning can perform clustering with fairly accurate accuracy may
still be extremely difficult to fully implement permanently. because even though the data we get comes from
customers and is structured, we are talking about humans, they can learn, and of course, changing a habit or
changing their spending patterns is something they might do. Assuming that implementing clustering like this
can give wrong results, it is safer to still let a manager make decisions in determining a target or strategy.
however, this does not close the answer that its application failed as the fact that the results we get in this
study can be arguably appropriate for use. The application of machine learning in this study may open up the
potential for other applications in the same industry
References
[1] C. Calvo-Porral and J. P. Lévy-Mangin, “Profiling shopping mall customers during hard times,” J. Retail. Consum.
Serv., vol. 48, no. November 2018, pp. 238–246, 2019, doi: 10.1016/j.jretconser.2019.02.023.

[2] A. G. Parsons, “Assessing the effectiveness of shopping mall promotions: Customer analysis,” Int. J. Retail Distrib.
Manag., vol. 31, no. 2, pp. 74–79, 2003, doi: 10.1108/09590550310461976.

[3] L. Lucia-Palacios, R. Pérez-López, and Y. Polo-Redondo, “Does stress matter in mall experience and customer
satisfaction?,” J. Serv. Mark., vol. 34, no. 2, pp. 177–191, 2020, doi: 10.1108/JSM-03-2019-0134.

[4] M. F. Diallo, F. Diop-Sall, S. Djelassi, and D. Godefroit-Winkel, “How Shopping Mall Service Quality Affects
Customer Loyalty Across Developing Countries: The Moderation of the Cultural Context,” J. Int. Mark., vol. 26, no.
4, pp. 69–84, 2018, doi: 10.1177/1069031X18807473.

[5] L. Kouhalvandi, O. Ceylan, and S. Ozoguz, “Automated Deep Neural Learning-Based Optimization for High
Performance High Power Amplifier Designs,” IEEE Trans. Circuits Syst. I Regul. Pap., pp. 1–14, 2020, doi:
10.1109/tcsi.2020.3008947.

[6] S. Hidayat, M. Matsuoka, S. Baja, and D. A. Rampisela, “Object-based image analysis for sago palm classification:
The most important features from high-resolution satellite imagery,” Remote Sens., vol. 10, no. 8, 2018, doi:
10.3390/RS10081319.

[7] L. L. Rego, N. A. Morgan, and C. Fornell, “Reexamining the market share-customer satisfaction relationship,” J.
Mark., vol. 77, no. 5, pp. 1–20, 2013, doi: 10.1509/jm.09.0363.

[8] X. Luo and C. B. Bhattacharya, “Social Responsibility , Corporate Customer and Market Satisfaction , Value,” Am.
Mark. Assoc., vol. 70, no. 4, pp. 1–18, 2006.

[9] A. Doulamis and N. Doulamis, “Customer Experience Survey,” pp. 3–6, 2016, doi: 10.3390/technologies8040076.

Musthofa Galih Pradana and Hoang Thi Ha / JADS Vol. 2 No. 1 2021
Journal of Applied Data Sciences ISSN 2723-6471
Vol. 2, No. 1, January 2021, pp. 19-25 25
[10] Akmal, “Predicting Dropout on E-learning Using Machine Learning,” J. Appl. Data Sci., vol. 1, no. 1, pp. 29–34,
2020.

[11] J. P. Ruiz, J. C. Chebat, and P. Hansen, “Another trip to the mall: A segmentation study of customers based on their
activities,” J. Retail. Consum. Serv., vol. 11, no. 6, pp. 333–350, 2004, doi: 10.1016/j.jretconser.2003.12.002.

[12] O. Dogan, C. Fernandez-Llatas, and B. Oztaysi, Process mining application for analysis of customer’s different
visits in a shopping mall, vol. 1029. Springer International Publishing, 2020.

[13] K. jae Kim and H. Ahn, “A recommender system using GA K-means clustering in an online shopping market,”
Expert Syst. Appl., vol. 34, no. 2, pp. 1200–1209, 2008, doi: 10.1016/j.eswa.2006.12.025.

[14] LeHew, M. L., & Wesley, S. C. (2007). Tourist shoppers' satisfaction with regional shopping mall experiences.
International Journal of Culture, Tourism and Hospitality Research, 1(1), 82-96.

[15] Anselmsson, J. (2006). Sources of customer satisfaction with shopping malls: a comparative study of different
customer segments. International Review of Retail, Distribution and Consumer Research, 16(1), 115-138.

[16] McGoldrick, P. J., and Thompson, M. G. (1992). The role of image in the attraction of the out-of-town centre.
International Review of Retail, Distribution and Consumer Research, 2(1), 81-98.

[17] M. Imron, U. Hasanah, and B. Humaidi, “Analysis of Data Mining Using K-Means Clustering Algorithm for
Product Grouping,” IJIIS Int. J. Informatics Inf. Syst., vol. 3, no. 1, pp. 12–22, 2020, doi: 10.47738/ijiis.v3i1.3.

[18] Akmal, “Predicting Dropout on E-learning Using Machine Learning,” J. Appl. Data Sci., vol. 1, no. 1, pp. 29–34,
2020, [Online]. Available: http://bright-journal.org/Journal/index.php/JADS/article/view/6.

Musthofa Galih Pradana and Hoang Thi Ha / JADS Vol. 2 No. 1 2021