Downloaded from https://iranpaper.

ir
https://www.tarjomano.com https://www.tarjomano.com

2019 2nd International Conference on Safety Produce Informatization (IICSPI)

The Application Research of Customer

Segmentation Model in Bank Financial Marketing

Yongxiang Feng Xiaoxin Wang Leixiao Li

School of Data Science and School of Data Science and School of Data Science and
Application Application Application
Inner Mongolia University of Inner Mongolia University of Inner Mongolia University of
Technology Technology Technology
China China China
[email protected] [email protected] [email protected]

Abstract—With the rapid development of information management of customer classification information, the
technology and internet finance, RFM model technology has model can reach the purpose of formulating targeted
been widely used in banking financial services. The traditional marketing strategy for the marketing personnel. The system
financial services industry has gradually changed from is recommended by the customers in accordance with the
product center to customer center. As customer transaction grade to achieve the purpose of marketing guidance. The
data continues to grow in the database, it is urgent to analyze RFM model does not require a specific value calculation for
the transaction information with the high efficiency big data the customer lifetime value (CLV), so it is more practical
analysis technology. Therefore, an RFM model suitable for [5].
financial customers has been established, and this model has
been applied to the financial product recommendation Three of RFM variables in determining the specific
guidance system. It can greatly improve the quality of calculation method, which usually standardized processing.
customer marketing services in the banking industry, and can After processing, the data will be calculated or clustered
effectively reduce operating costs. [6,7]. The concrete process will be conducted in the next
chapter in detail. First of all, the information of original data
Keywords—RFM model, customer segmentation, prepossessing to get score, then the rating information for K
recommendation guidance - means clustering, finally analyze the clustering results and
I. INTRODUCTION get the customer classification information, this is the way of
FRM model building.
At the beginning of the 21st century, China joined the
WTO, which promoted the further development of the The RFM model on data processing, operate more easily
domestic financial market pattern. Financial markets are than other models, the results of the analysis is also easier to
enjoying a new boom [1].The prosperity of financial market understand [8]. Based on 144109 transaction records of
makes between traditional banking and all kinds of new type 3,7170 clients, this paper deals with the transaction data of
of financial services industry to customers increasingly financial clients of commercial Banks. According to RFM
fierce [2]. The traditional banking industry relying on the model, data prepossessing is conducted and customer
inherent mechanism consists of various branches of the segmentation model is established. The model uses the
passive customer marketing, which can’t satisfy the needs of distribution of R, F and M score values to classify customers,
customers and can’t provide customers with satisfactory and finally analyzes the customers by using the results
service [3]. In recent years, wealth management business has obtained by clustering.
been developing rapidly in traditional banking. As a result, B. RFM Model in Financial Customer Segmentation
many Banks have turned their wealth management
businesses into key businesses. The development of personal Although RFM model is a common method for customer
finance business has promoted the marketing reform of segmentation, RFM indicators need to be optimized and
traditional banking industry. As the core to the customer’s improved according to specific conditions. In order to apply
core market transformation, financial business marketing to the bank financial clients transaction information data
services are also changes in the traditional Banks.t has analysis [9]. If established in accordance with the traditional
become an undisputed consensus to provide customers with model of RFM indicators will appear many problems, for
appropriate level services and to provide high-quality, example, the last trading date only statistical customer with
efficient and targeted banking financial marketing the number of days interval, testing benchmark dates will
recommendations to customers [4]. not be able to analyze the bank financial trading time span of
the customer, unable to distinguish between old and new
II. RELATED THEORIES customers at the same time; For example, a customer in a
certain period of frequent to the operation of the open to
A. Application of RFM Model for Financial Clients redeem to buy financial products, trading frequency index
Based on the analysis of the transaction information will continue to accumulate, unable to accurately analyze the
behavior of financial clients, the RFM model of customer customer segmentation frequency standards. Since the
value was created. The model is mainly applied to the amount of investment is also an accumulative value, it will
classification of customers. According to the system result in a very large accumulative value in the transaction of

978-1-7281-4566-2/19/$31.00 ©2019 IEEE -564-

Authorized licensed use limited to: UNIVERSITAET BAYREUTH. Downloaded on November 09,2024 at 07:08:59 UTC from IEEE Xplore. Restrictions apply.
 Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

2019 2nd International Conference on Safety Produce Informatization (IICSPI)

frequent redemption and repurchase, which is obviously will process the detailed transaction data of the customer, the
unreasonable. So the traditional RFM model index cannot necessary system user data and customer information data,
apply to bank financing of customer segmentation. To this and then enter into the system.
end, the paper puts forward using customer enclosed
products trading information as data base, and quarterly (2) Data Preprocessing
average investment according to the customer for data 1) Data Cleaning
analysis, and then established is suitable for the financial The main purpose of data cleaning is to deal with invalid
RFM model of customer segmentation. The traditional RFM data in the data, and the scope of invalid data is directly
index and improved RFM index for such as shown in Table related to the selection of the model and the property of the
I. transaction data itself.
TABLE I. THE TABLE OF COMPARISON OF INDEX MODEL The invalid data caused by the selection of the model in
this article mainly includes: the transaction attributes
Traditional In this article, the financial required for non-RFM models can be removed directly. At
Model indexes
RFM index customer RFM index
For the first time in a certain period
the same time, the missing data corresponding to the three
(two years), the time interval values of proximity, frequency and value should be filled
between the distance analysis points and processed.
The time interval
of the closed financial products is
between the Invalid data in transaction information is mainly includes:
R purchased.
customer’s last the failed transaction in the transaction data, the remote data
(Proximity) The time interval between the
trading distance test
point
distance analysis points of the closed and products of distance model time, and the data of
financial products is the last time the non-closed type.
customer purchases the closed
financial products. 2) Data Specification
The average number of times The primary purpose of the data attribute specification is
The total number of (quarterly frequency) of each type of to delete the attributes that are not related to the model in the
F
(Frequency)
times a client trades financial products purchased by original data. For example, the entry teller, customer name
in a period of time. customers within a certain period and product dividend method in the transaction information.
(two years). Select the valid attribute for the data set through the data
attribute specification: customer number, transaction
confirmation date, transaction confirmation amount and
The total amount of During the period (two years), the transaction number.
M money that the customer will purchase the amount
(Monetary) customer deals for a of the closed financial products every 3) Data Conversion
period of time. quarter.
The main purpose of data conversion is to unify the
different measurement units in raw data. Because transaction
information data is processed, the transaction date,
III. CONSTRUCTION AND SYSTEM APPLICATION OF
transaction frequency and transaction amount are different,
FINANCIAL CUSTOMER SEGMENTATION MODEL standardization is required before the model construction is
A. Research Content carried out.
Firstly, the improved RFM model is used to model the (a) Standardized processing of data in the near
customer segmentation, and then the Oracle relational degree
database is used to store the original data of the model. The In this paper, the near degree values of RFM model are
system uses the Linux language editing system script to set standardized. The process is to record the first transaction
the data transmission of the Oracle relational database and date Rf and the last transaction date Rl, i.e. the date of the
the HDFS distributed file system. The system uses Scoop transaction after the first transaction and the last transaction
data transmission technology as the transmission bridge confirmed by the customer.
between two databases. Secondly, the system uses the
machine learning library in Spark On Yarn platform to The data in the paper comes from the real data provided
optimize k-means clustering technology. Through the by a city commercial bank. The data provided from the
combination of data and data mining technology, the beginning of 2011 to the end of mid-April 2017. This paper
customer data can be analyzed effectively and the valuable selects the data from April 2015 to March 2017 as the initial
information of the customer can be deeply mined. Finally in data set of the model, and the observation date is selected as
the system through the clustering results of customer the last day of April 2017.
segmentation model of display, and analyze the content
During April 2015 to March 2015 customer transaction
classification information. To bring more opportunities
data according to the trading time for positive sequence
recommended guidance for traditional banking, these
arrangement, which can be counted in the time range of the
opportunities are buried in the customer transaction
customer’s trading for the first time the date and time at the
behavior.
end of the transaction date. The time interval of this paper is
B. Preparation of Data two years, and there are eight quarters in total, which can be
divided into eight scores to evaluate the difference, which
(1) Data Acquisition
fully reflects the difference of the importance of proximity in
The initial data is statistics from the bank’s financial
predicting customer transaction behavior. Therefore,
trading system, the system after a unified text format
according to the customer’s transaction confirmation date,
specification, system to save the data text, thus ensuring the
the distance from the observation time is 1, and the recent
data consistency. The bank’s financial transactions system

-565-
Authorized licensed use limited to: UNIVERSITAET BAYREUTH. Downloaded on November 09,2024 at 07:08:59 UTC from IEEE Xplore. Restrictions apply.
 Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

2019 2nd International Conference on Safety Produce Informatization (IICSPI)

rule of 8 is graded. Furthermore, the corresponding values of the total amount of the cumulative value of the last
Rf and Rl are evaluated as Rfs and Rls. The specific Rf and transaction, Rfs for the first time trade score value, Rls for
Rl processes are shown in Table II. the last transaction score value.

TABLE II. THE TABLE OF RECENCY DATA NORMALIZATION. Mq M /[( Rls  Rfs  1) x F ] . (2)
Customer Share In this paper, the actual transaction data and product data
Start Date End Date Score Evaluation
(%)
2015-04-01 2015-06-30 1 1.821
of the bank are selected. The minimum amount of the bank’s
2015-07-01 2015-09-30 2 5.736 trading votes for fifty thousand yuan, so there will be no
2015-10-01 2015-12-31 3 7.474 average transaction amount is less than 50000.
2016-01-01 2016-03-31 4 13.352
2016-04-01 2016-06-30 5 14.563
Value of standardized treatment in this article also uses
2016-07-01 2016-09-30 6 16.083 the score evaluation method, the method of average
2016-10-01 2016-12-31 7 26.478 transaction rate statistics, and the similar customer
2017-01-01 2017-03-31 8 14.493 purchasing behavior as together, the score evaluation of the
(b) Standardized processing of frequency data value of Mqs. As shown in Table IV.
Average customer purchase frequency Fq is a customer TABLE IV. THE TABLE OF VALUE DATA NORMALIZATION
during April 2015 to March 2017, the number of customers
per quarter investment enclosed products, is the customer in Customer
Initial amount End of The Amount Benchmark score
this transaction successful quarter average number within Share (%)
two years. The standardized treatment is shown in Eq. 1. 50000.0 80000.0 1 19.066
Which Fq for the quarterly average number, F for the 80000.0 120000.0 2 15.623
customer from the transaction date for the first time to the 120000.0 210000.0 3 19.015
last trading date between investment in the overall number 210000.0 310000.0 4 12.604
of statistics, score value Rfs for the first transaction, Rls for 310000.0 452000.0 5 10.665
the last transaction score value. 452000.0 720000.0 6 10.223
720000.0 1380000.0 7 7.805
Fq F /( Rls  Rfs ) . (1) 1380000.0 414660000.0 8 4.999

Because of the actual data will be the same customer at C. System Construction and Application of Customer
the same day of the incremental transactions, the situation Segmentation Model
will be the same day many times as a trading, the amount of The construction and application of customer
investment will be graded on statistics, the aggregate amount segmentation model are mainly divided into two main
of the Fq of 1. components:
In this paper, the value of the quarterly average number The first part is the data analysis of customer
Fq is evaluated by using the previous score evaluation segmentation. System using historical customer purchasing
method, which is the same as Rf and Rl. As far as possible, behavior information for data analysis, and then the system
the data of similar frequency value should be classified into through the pretreatment of RFM analysis score customer’s
a standard, the frequency of score evaluation for Fqs, transaction database, Then, the system uses the value score
specific Fq according to the score evaluation standard for of the three attributes of RFM analysis to express the
processing the list below, as shown in Table III. customer, and finally the customer cluster analysis is
conducted.
TABLE III. THE TABLE OF FREQUENCY DATA NORMALIZATION.
The second part is the model application in the financial
Customer
Starting Frequency End Frequency Benchmark Score
Share (%)
product recommendation guidance system. According to the
0 1 1 51.308
customer clustering information analyzed in the first part,
2 3 2 45.254
the customer is analyzed by the system to obtain the
4 5 3 3.037
customer classification information. And combined with the
customer’s risk tolerance, the marketing personnel to carry
6 7 4 0.350
on the sale of financial products recommendation guidance.
8 9 5 0.035
10 11 6 0.008 (1) Construction of RFM Customer Segmentation Model
12 13 7 0.003 Results the collection point diagram is shown in Fig. 1.
14 15 8 0.005
As can be seen from Fig. 1, when K=9, the calculation
(c) Standardized processing of value data results fluctuate, but after the fluctuation, it tends to be stable,
The average customer transaction amount Mq is the so we choose 8 as the value of K.
average value of the client’s investment in financial products
transactions between April 2015 and March 2017.Client The specific process of the algorithm is shown in Table
average transaction amount Mq is a customer in April 2015 V.
to March 2017 investment products trade value, the average In the process of k-means clustering, the selection of the
amount of value is the amount of time interval average total initial clustering center point of step 3 will result from the
amount of the investment to the value on each transaction in different influence of the center point.In the k-means of
the each quarter. The standardized treatment of values is Spark MLlib, there is a mode “Kmeans||” which can be
shown in Eq. 2, which Mq for the average transaction selected for the initial cluster center, follow the basic
amount, M for the period from trading for the first time to principles of K-means++ algorithm of this model, the initial

-566-
Authorized licensed use limited to: UNIVERSITAET BAYREUTH. Downloaded on November 09,2024 at 07:08:59 UTC from IEEE Xplore. Restrictions apply.
 Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

2019 2nd International Conference on Safety Produce Informatization (IICSPI)

clustering center of the distance between each other should K

be as far as possible,.You also can intuitively understand that
this is going to be better for the K initial cluster centers.
SSE ¦ ¦ Edist (w , x)
i 1 xWi
i
2

. (3)
TABLE V. THE TABLE OF ALGORITHM STEPS. In the process of k-means clustering, the fifth step needs
Algorithm Step Description
calculate average value of all points in each cluster, the
1 Read the RFM analysis score data set W. calculation steps are as follows.
2 Determine and select K values for clustering.
It can be seen from Eq. 3 that the optimal center of mass
3 Initial K cluster center point selection.
4 Calculate the Euclidean distance between the data
can be obtained when the error sum of squares is the lowest.
concentration points and K clusters, and select the nearest So the formula is minimized, and the derivative of S with
central point to divide into the cluster. respect to S is equal to zero, as shown below:
5 Calculate the mean value of the coordinates of all points
K
in each cluster as the new center. w w
6 Whether the algorithm satisfies the convergence SSE ¦ ¦ ( w , x) 2

w w
i
condition S, if satisfied to step 8, does not satisfy the Wk wk i 1 xWi
convergence condition to step 7. K
w
7 Determine whether the algorithm reaches the maximum
¦ ¦ ( w , x) 2
. (4)
w
iteration threshold, if it reaches the eighth step, and does i
not reach the step 4, 5 and 6 of the cycle. i 1 xWi wk
8 Output K cluster corresponding result set.
¦ 2(w  x ) 0
xWk
k k

¦ 2(w
xWk
k  xk ) 0 Ÿ n k wk ¦x
xWk
k

1
Ÿ wk
nk
¦x
xWk
k
. (5)
In other words, for wi, when you take the derivative,
wk
1
nk ¦x
xWk
k

when , the SSE minimizes the SSE’s optimal

center of mass is the mean of each point in the cluster.
Therefore, the formula can be deduced similarly. For xi,
when xi=wi, SSE is the smallest, and corresponds to step 4
in the process of k-means clustering.

Fig. 1. The Scatter plot of the K selected results. In conclusion, as the number of iterations increases, the
SSE value decreases to a minimum value, so SSE converges.
The basic steps are as follows: According to the clustering results, the spatial scatter
diagram is drawn to show the clustering distribution as
1) Randomly select a point from the data set X as the shown in Fig. 2.
first starting point and define X as a sample of X set.
2) Calculate the shortest distance D(x) between each x
and the cluster center.
3) Calculate the probability of each x being selected as
the next class cluster center

D( x) 2
P( x)
¦ D( x)
x X
2

;
Choose the largest x of P(x) as the next cluster center.
(2) Repeat the above two steps until the end of the K
initial center selection.
In the process of k-means clustering, the convergence
condition S is satisfied in step 6, which is usually used to Fig. 2. The Scatter plot of clustering distribution.
determine whether the discriminant condition of the
convergence of K-means algorithm is the squared error and (3) Model Application in Financial Product
S, and its definition is as shown in Eq. 3. Where K Recommendation Guidance
represents K cluster centers, wi represents the focal point of Customer segmentation model in financial products
the first i cluster center, and Edist represents Euclidean recommended the application of the guidance system is
distance. mainly, through the analysis of customer classification
segmentation results to customers recommend products, at
the same time, system can provide a guideline for marketing

-567-
Authorized licensed use limited to: UNIVERSITAET BAYREUTH. Downloaded on November 09,2024 at 07:08:59 UTC from IEEE Xplore. Restrictions apply.
 Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

2019 2nd International Conference on Safety Produce Informatization (IICSPI)

personnel. First, the model represents the RFM score of the commercial banks’ financial products.
customer. Next, the model cluster analysis to the customer
and the results of the clustering are matched. Finally, the Clustering group 7 is important call customers. In the
customer classification information is obtained. This is the RFM model of customer subdivision of bank financial
specific application process of the model in financial product product recommendation guidance, the freshness of
recommendation guidance system. The system matches the investment is lower than the average value but closer to the
of the customer and the hierarchical information of the average value. The consumption frequency and the average
customer’s risk tolerance, and then carries out the product investment amount of each quarter are higher than the
recommendation according to the matching information. The average. These clients are not very interested in recent
specific process is shown in Fig. 3 below. investments, but having frequent and high investments, so
they are important call customers.
begin
TABLE VI. THE TABLE OF ANALYSIS RESULTS
Average Average
Enter Customer does not Exist Class Average of Customer Compare Customer
of of
Customer Number Please Re-enter Number Frequency Number Results Level
Recency Monetary
Important
1 7.16 1.71 7.39 3870 5Ĺ)Ĺ0Ĺ
customer
Transaction Account General
No
Exists 2 4.64 1.59 3.07 4506 5Ļ)Ĺ0Ļ maintenance
of customer
Yes Loss of
3 3.67 1.37 1.30 6028 5Ļ)Ļ0Ļ
Customer Risk customer
Tolerance Risk Assessment Potential
No 4 6.89 1.55 3.45 6661 5Ĺ)Ĺ0Ļ
Assessment customer
General
Yes 5 6.84 1.38 1.47 4882 5Ĺ)Ļ0Ļ observation
of customer
Classification Loss of
Information Exists
No 6 2.05 1.29 2.75 2655 5Ļ)Ļ0Ļ
customer
Call back
Yes 7 4.23 1.61 6.10 2774 5Ļ)Ĺ0Ĺ the
customer
Query Customer’s Risk Level Query Customer’s Important
8 7.05 1.68 5.50 5794 5Ĺ)Ĺ0Ĺ
and Classification Information Risk Level customer
Total 5.60 1.53 3.67 37170
(5) Information Analysis of Financial Customer
Choose the Right Segmentation Model
Product No It is known from Table VI in the previous section, cluster
Recommendation
1 and cluster 8, two customer groups are important value
Yes Query Shows Inappropriate
customers. Their RFM model index of customer
Product Information segmentation in the bank financial product recommendation
Query Shows Suitable
Product Information
guidance system is higher than the average. It shows that the
recent investment of such clients is relatively close, the
number of investment per quarter is higher, and the average
End investment amount per quarter is higher. This kind of client
has a great contribution to the development of bank finance
Fig. 3. The flow chart of financial management product recommendation.
and has been formed in the bank to buy financial products
consumption habits. This reflects the recognition of
Based on the clustering results, the average value of the
commercial banks’ financial products.
RFM model score of the eight kinds of customers is
calculated. Then the system compares the average score Clustering group 7 is important call customers. In the
within the class with the average score of the whole. Finally, RFM model of customer subdivision of bank financial
the customer is redefined according to the comparison result product recommendation guidance, the freshness of
of system, and the results are shown in Table VI. investment is lower than the average value but closer to the
average value. The consumption frequency and the average
(4) Information Analysis of Financial Customer
investment amount of each quarter are higher than the
Segmentation Model average. These clients are not very interested in recent
It is known from Table VI in the previous section, cluster investments, but having frequent and high investments, so
1 and cluster 8, two customer groups are important value they are important call customers.
customers. Their RFM model index of customer
segmentation in the bank financial product recommendation Clustering group 2 is generally retain customers. In the
guidance system is higher than the average. It shows that the RFM model of customer subdivision of bank financial
recent investment of such clients is relatively close, the product recommendation guidance, the freshness of
number of investment per quarter is higher, and the average investment and average investment in the amount of each
investment amount per quarter is higher. This kind of client quarterly investment are lower than the average value but
has a great contribution to the development of bank finance closer to the average value, and the investment frequency is
and has been formed in the bank to buy financial products higher . The customers have certain investment habits. Each
consumption habits. This reflects the recognition of time they can control the amount of investment and consider

-568-
Authorized licensed use limited to: UNIVERSITAET BAYREUTH. Downloaded on November 09,2024 at 07:08:59 UTC from IEEE Xplore. Restrictions apply.
 Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

2019 2nd International Conference on Safety Produce Informatization (IICSPI)

the investment cycle, so the customers need to maintain. investment products integrated trading behavior, and then
analysis the result, will eventually analysis results and the
Clustering group 4 is the potential customers. In the integration of marketing strategy to establish financial RFM
RFM model, which is recommended by bank financial model of customer segmentation. On the basis of
products, the investment freshness and the quarterly guaranteeing stability, the system improves the analysis
investment frequency are higher, but the average investment efficiency, so that the customer classification analysis and
amount per quarter is lower than the average. The clients the bank financial products recommendation and guidance
have been investing heavily in the near future, but they can management achieve effective combination. Therefore, this
control the amount of each investment. Therefore, the clients system has made an important contribution to the bank’s
need to tap the investment potential. further promotion of marketing service level and the
Clustering group 5 is general observation customers. In exploration of customers’ potential ability.
the RFM model, which is recommended by bank financial
products, the investment freshness is higher. But the REFERENCES
quarterly investment frequency and the average investment [1] J.Y. Bao, “On Global Financial Impact, Market Constraint and
amount per quarter is lower than the average. This kind of “Hostile risk” Governance of China’s “One Belt And One Road”
Overseas Investment”, China Soft Science, vol. 3, pp. 1-10, 2017.
customers may be for short-term users of low investment
[2] X. Ruan, “Design and Implementation of Customer Relationship
and new customers of bank financial products which has the Management System of Guiyang Wudang Bank”, Xiamen: Xiamen
certain development space, so belong to need to continue to University, 2015.
observe to the investment potential. The customers are called [3] N.G. Mu, H. Wang, “Opportunities and Challenges in the
general value customers. Development of Traditional Commercial Banks under the
Background of Internet Finance”, E-commerce, vol. 2, pp. 36, 2016.
Clustering group 3 and 6 are the loss of our customers. In [4] M. Dietz, P. Harle, S. Khanna, et al., “Fighting for Customers: How
the RFM model, which is recommended by bank financial Traditional Banks Deal With Internet Financial Challenges”, New
products, the investment freshness, the quarterly investment Finance, vol. 1, pp. 37-44, 2017.
frequency and the average investment amount per quarter is [5] Y.G. Feng, “Empirical Analysis on the Value of the Customer’s
lower than the average. So they need to be left behind. Word-of-mouth Recommendation based on Customer Lifetime
Value”, Statistics and Decision-making, vol. 12, pp. 186-188, 2016.
IV. CONCLUSIONS [6] X.Q. An, “Research on Customer Value Recognition based on RFM
Model in C2C Environment”, Chengdu: Southwestern University of
The system uses the improved RFM index to conduct Finance and Economics, 2012.
data preprocessing to achieve the purpose of analyzing the [7] L. Xiong, B. Gao, “Research on Customer Segmentation based on
transaction information behavior of financial clients. At the RFM Multi-level Customer Value Model”, Business Era, vol. 5, pp.
same time, the system normalizes the original data of the 55-57, 2017.
model, and then the system uses the k-means clustering [8] H.L. Li, “Application of Data Analysis and Mining in the Refined
technique commonly used in data mining to excavate the Operation of E-commerce”, China Economic and Trade, vol. 15, pp.
original data of the model. Then the system is analyzed 52-53, 2017.
according to the clustering results to obtain the customer [9] X.X.Wu, “User Segmentation of Internet Finance Platform based on
classification information. According to the real bank RFM Improvement Model”, Beijing: Beijing Jiaotong University,
2016濁
financial customer transaction data, based on the closed-end

-569-
Authorized licensed use limited to: UNIVERSITAET BAYREUTH. Downloaded on November 09,2024 at 07:08:59 UTC from IEEE Xplore. Restrictions apply.