See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/323834276

Detection of money laundering accounts using data mining techniques

Conference Paper · March 2018

CITATIONS READS
0 3,846

2 authors:

Priyanka Sain Shalini Puri

Poornima Group of Colleges Manipal University Jaipur
1 PUBLICATION 0 CITATIONS 39 PUBLICATIONS 239 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Shalini Puri on 02 June 2019.

The user has requested enhancement of the downloaded file.

 Detection of money laundering accounts using data mining techniques
Priyanka Sain
Poornima College Of Engineering, Jaipur, Rajasthan, India
Email: [email protected]
Shalini Puri
Poornima College Of Engineering, Jaipur, Rajasthan, India

Abstract – try to convert monetary proceeds derived from illicit

Today, money laundering poses a serious threat not only to activities into “clean” funds using a legal medium such as
financial institutions but also to the nation. Money large investment or pension funds hosted in retail or
laundering is a criminal activity to disguise black money as investment banks. Today, ML is the third largest
white money. It is a process by which illegal funds and assets “Business” in the world following the Currency Exchange
are converted into legitimate funds and assets. Money and the Automobile Industry. One can only forecast
Laundering occurs in three stages: Placement, Layering, and according to the fraud that has already been exposed.
Integration. It leads to various criminal activities like Nowadays, it poses a serious threat not only to financial
Political corruption, smuggling, financial frauds, etc. In institutions but also to the nation. Some risks faced by
India there is no successful Anti Money laundering financial institutions can be listed as reputation risk,
techniques which are available. To spot potential clusters of operational risk, concentration risk and legal risk. At the
criminals, different methods were proposed and we will society level, ML could provide the fuel for drug dealers,
study all the methods proposed and evaluate the accuracy terrorists, arms dealers and other criminals to operate and
and best techniques that can be used for performing expand their criminal enterprises. Therefore, anti-money
detecting suspicious accounts. Recently, data mining laundering (AML) is of critical significance to national
approaches have been developed and are considered as well- financial stability and international security. Traditional
suited techniques for detecting money laundering activities. investigative techniques consume numerous man-hours.
In this paper, we present different methodologies used for Traditional approaches to AML followed a labor intensive
detecting suspicious accounts involved in money laundering manual approach because ML is a sophisticated activity
activities. We present a data mining approach for profiling with many way of laundering money. These approaches
bank clients in order to support the process of detection of can be classified into the identification of money
anti-money laundering operations. We detail the laundering incidences, detection, avoidance and
experiments performed on real world data from a financial surveillance of money laundering activities [1]. Indeed,
institution, which allowed us to group clients in clusters and given that the volume of banking data and transactions
then generate a set of classification rules. We discuss the have increased in number of ways, such approaches need
relevance of the founded client profiles and of the generated to be supported by automated tools for detecting money
classification rules. According to the defined overall agent- laundering’s pattern. Meanwhile, AML software tools in
based architecture, these rules will be incorporated in the the market are normally rule-based that make the decisions
knowledge base of the intelligent agents responsible for the using some sets of predefined rules and thresholds based
signaling of suspicious transactions. on mean and standard deviation values. Besides, data
mining techniques (DM) [2] have been proven to be well
Index Terms—Data mining, Anti Money Laundering, suited for identifying trends and patterns in large datasets.
clustering, neural networks, classification; customer Therefore, DM techniques are expected to be applied
clustering; multi agent systems; suspicious transactions successfully in the area of AML. Nevertheless, there is still
little research concerning this bias especially a DM
I. Introduction framework/solution for supporting AML experts in their
daily tasks. Recently, there are some AML approaches
Money laundering is a process of converting unaccountable based on DM that have been proposed and discussed in
money in to accountable money by making illegitimate literature. Most of these approaches try to recognize ML
income appear legitimate. This criminal activity leads to patterns by different techniques such as support vector
various adverse effects ranging from drug trafficking to machine [3], correlation analysis [4], histogram analysis
financial terrorism. Through money laundering, criminals [4], etc. They aim to provide techniques for detecting a
 variety of ML by exploring a massive dimensionality of and results of this approach are presented and discussed
datasets including customers x accounts x products x and in section 5 the conclusion and Further Enhancement
geography x time. However, these approaches are more or has been explained.
less appropriate for the cash world and not scaled well for
investment activities due to the lack of good methods in
choosing parameters and they still have performance issues. II. Literature Review
Customer behavior in investment activities is complicated
because it is influenced by many factors. We also show that Nhien An Le Khac et al. [5][6][7] constructed a data
by choosing suitable parameters, simple DM techniques can mining based solutions to examine transactions for
be applied together to detect suspicious ML cases in detecting money laundering and also suggested an
investment activities. Data Mining is an area in which huge investigating process based on different data mining
amounts of data are analyzed in different dimensions and techniques such as Decision tree, genetic algorithm and
angles and further categorized and then eventually fuzzy clustering. By merging natural computing
summarized in to useful information. Data Mining is the techniques and data mining techniques; they proposed a
process of finding correlation or patterns among dozens of knowledge based solution to detect money laundering.
fields in large databases. The three stages of money Different approaches were proposed for quick
laundering include Placement, Layering and Integration. identification of customers for the purpose of application
The placement stage is the stage where in the actual criminal of Anti money laundering. In their paper implemented an
person disposes all the illegal cash to a broker. This broker approach wherein they determined the important factors
or agent is responsible for distributing money. In the layering for investigating money laundering in the investment
stage the cash is spread into multiple intermediaries that can activities and then proposed an investigating process based
include banks and other financial institution. The major issue on clustering and neural network to detect suspicious cases
lies in this layering stage of money laundering because here in the context of money laundering. Jiawei Han et al. [8]
the transfer of money may be from one to one or one-to- in their book focused on improving the efficiency of
many .The difficulty arises in tracing out all the chaining of apriori algorithm and then explained Hash based technique
transactions. In the integration stage all the cash is to reduce the size of candidate k –tem set, Ck, for k>1. [9]
transferred to a beneficiary often called as Integrator. At this applied rule-based evaluation to identify suspicious
stage all the transactions are made legal. To trace out the parties. The rules used by this system were derived from
dirty proceeds immediately this proposed framework aims at expert knowledge and encoded in an evaluation module
developing an efficient tool for identifying the accounts, that executed each time the target database was updated.
transactions and the amount involved in the layering stage of Parties matching these rules would then investigated by
money laundering. analysts using an interactive query interface and
visualisation tools provided by the system.
The rest of the paper is organized as follows; the literature
survey is presented in section-2 of the paper, section3 of the
paper deals with the proposed method and provide some
context for our work. Section 4 the experimental analysis

III. Proposed Work

Reference No. Author Problem Domain

[10] Liu Keyan, An improved Support vector Network Model for Anti-Money Laundering

Yu Tingting

[11] Assem Khalaf Ahmed, Data Mining Techniques for Anti Money Laundering

2
 Allam El-Din,

Nashaat El Khamesy

[12] Mahesh Kharote, Data Mining Model for Money Laundering Detection in Financial Domain

V. P. Kshirsagar

[13] Andrea Fronzetti Using social network analysis to prevent money laundering

Colladon,

Elisa Remondi

[14] Z. Zang, Applying Data mining in Investigating Money Laundering Crimes

J.J. Salermo,

P. S. Yu

[15] Ahmad Salehi, Data Mining Techniques for Anti Money Laundering

Mehdi Ghazanfari,

Mohammed Fathian.

[16] Zengan Gao and Mao A framework for data mining-based anti-money laundering research
Ye

[17] R.Chandni, A Graph Based Approach to Identify the Suspicious Accounts by

Implementing the Hash Based Association Mining
G. Deepa Sakthi,

B.Geerthana,

A.Rajalakshmi

[18] Ch. Suresh, A Hybrid Approach for Detecting Suspicious Accounts in Money
Laundering Using Data Mining Techniques
Dr. K. Thammi Reddy,

N. Sweta

3
[19] Nhien-An Le-Khac, A Heuristics Approach for Fast Detecting Suspicious Money Laundering
Sammer Markos and Cases in an Investment Bank
M-Tahar Kechadi

[20] Alexandre, Claudio Client Profiling for an Anti-Money Laundering System

Balsa, Jo˜ao

Table 1. list of methods viewed and evaluated

Reference No. Classifiers Used Results Obtained Limitations

[10] support vector machine higher degree of precision lack of good

methods in
choosing
parameters

[11] neural networks, 1. accurate and fast 1. less

decision making approapriate
classification, 2. MLP provides the
2. lack of good
knowledge removal, best classification rate methods in
3. LNN neural network choosing
data pattern analysis gives: parameters
3.1. the best
evaluate four types of neural networks: 3. inefficiency
performance of
with large data
1. Multi-Layer Perceptron Neural Network (MLP) classification rate
sets can only
3.2. high performance of identify
2. Probabilistic Neural Network (PNN) training, selecting suspicious
and testing the data behaviour that is
3. Radial Basis Function (RBF)
3.3. lowest error similar to that
4. Linear Neural Network (LNN). observed in
previous
Radial basis function (RBF) networks investigations.

higher degree of precision

[12] 1. Transaction analysis improves the complexity performance
2. Clustering by reducing the clustering issues
3. frequent pattern mining problem to a segmentation
4. suspicious transaction problem
5. suspicious customer
6. visualization.
7. Transaction Flow Analysis (TFA)
8. Apriori-based GSP algorithm.

4
 9. clusters would be examined using Extracting
Customer Behavior
10. prefix-projection heavily reduces the size
of the projected data and leads to efficient
processing.
11. clusters the records into several sectors,
and then finds the correlation among these
sectors
high correction rate
[13] 1. Network Analytic Techniques reducing false positive rate 1. inefficiency
2. network-based approach with large data
enhancing detection rate
3. Social network analyisis sets can only
a high processing speed identify
4. RuleBased Classification approach satisfactory accuracy suspicious
5. ontology based expert system behaviour that is
6. Bayesian approach similar to that
7. fuzzy techniques observed in
8. sort and map relational data previous
9. predictive models: network metrics investigations.
predict risk profiles using social network
metrics
10. spot potential clusters of Criminals
11. visual analysis of the tacit links
existing among different companies
who share the same owner or
Representative
12. Relational approach
13. Study of relationship networks
14. Created a binary dependent variable
that has the value of one if either the
owner or the representatives were
involved in legal trials and has the
value of zero otherwise.

[14] 1. correlation analysis improves the complexity by 1. less

2. histogram analysis reducing the clustering appropriate
3. exploring a massive dimensionality of problem to a segmentation 2. lack of good
problem
datasets including methods in
4. customers x accounts x products x choosing
geography x time parameters
3. performance
issues

Increased accuracy
[15] 1. unsupervised data mining technique Increased performance
2. detect new patterns of money laundering

5
 3. enhance learning models based on
classification method

[16] 1. link analysis deal with high dimensionality 1. time-

2. community generation heterogeneous data sets. consuming
3. network destabilization 2. resource
4. network analysis intensive
5. decision tree 3. high false
6. Bayesian inference positive rate
7. regression (FPR)
8. case-based reasoning 4. inefficiency
9. support vector machine with
voluminous
data sets.
5. random
fluctuations
because of
data
inconsistency

improved efficiency
[17] 1. hash based technique high correction rate performance
2. graph theoretic approach reducing false positive rate issues
3. time variant approach using behavioral enhancing detection rate .
patterns to identify money laundering a high processing speed
4. transaction logs split into various time satisfactory accuracy
window and for each account specific to the
fund transfer the time value is split into
different time windows
5. And it generate the behavioral pattern of the
user’s.
6. The behavioral pattern specifies the method
of transfer between the accounts. Based on
the generated behavioral pattern it identifies
the suspicious account.
7. Frequent transactional data sets were
collected and the traversal path of the
suspicious transactions were identified using
graph theoretic approach
8. The traversal path gives the illegal
transactions between the agent and the
integrator.
1. effectively trim the
[18] 1. Hash based Association approach transaction database size less approapriate
2. Graph Theoretic Approach.

6
 3. mining association rules at a much earlier stage of lack of good
4. identify the traversal path of the Laundered the iteration methods in
money 2. Reducing the choosing
parameters
5. identifying agent and integrator in the computational cost.
layering stage of Money Laundering 3. improves efficiency
6. finding correlation or patterns among dozens 4. successful in finding the
of fields in large databases agent and integrator in
7. construction of candidate set of item sets first the transaction path
identifying within this item set those item sets
that need the large item set requirement
8. The suspicious accounts of the layering stage
of the money laundering process are
identified by generating frequent
transactional datasets using Hash based
Association mining.
9. The generated frequent datasets will then be
used in the graph theoretic approach
10. to identify the traversal path of the suspicious
transactions
1. satisfies the needs of the
[19] 1. heuristics approach to improve the AML unit. System
performance for this solution 2. It can also improve efficiency not
2. combination of clustering and classification sufficient
significantly the
techniques for analysing ML patterns in an performance
international investment bank.
3. clustering algorithm is repetitively executed
to analysis transactions depending on the
characteristic of each transaction datasets.
4. one-step clustering approach basing on some
heuristics from AML experts to improve the
performance
5. divide the datasets into two kinds of
investors: corporate and individual as their
investment behaviours are different
6. two heuristics: suspicious screening and
suspicious initial centres to detect suspicious
cases
1. simplicity, efficiency 1. for
[20] 1. group clients in clusters and then generate 2. present in almost nominal
a set of classification rules every platform that attributes this
2. form groups of clients with similar implements and automates calculation can’t
characteristics and mutually exclusive the data mining process. be made.
3. K-means algorithm 3. obtain better results
when dealing with

7
 4. uses squared Euclidean distance to continuous numerical
calculate proximity attributes in comparison
with nominal attribute use
4. produced good
results for both the cluster
evaluation metrics and
generated rules.
5. build clusters that
represent risk groups and
rules both more
discriminating and with a
wider coverage, regarding
the profiles’ attributes
Table 2. methodologies used, results and limitations
2. Conclusions [2] J. Han and M. Kamber, Data Mining: Concept and
Techniques. Morgan Kaufmann publishers, 2nd Eds., Nov.
The proposed system examines the efficiency of the existing 2005.
anti money laundering techniques by identifying the [3] J. Tang, J. Yin, Developing an intelligent data
suspicious accounts in the layering stage of money laundering discriminating system of anti-money laundering based on
process These accounts have the highest possibility of being SVM, Proceedings of the Four International Conference
suspicious as there are involved in huge amount of on Machine Learning and Cybernetics, Guangzhou, Aug.
transactions frequently. The solution proposed here is highly 2005. pp. 3453-3457.
advantageous over the existing anti-money laundering rules. [4] Z. Zang, J.J. Salermo and P. S. Yu, Applying Data
Further enhancement: With the chaining of accounts, we can mining in Investigating Money Laundering Crimes,
further develop a system which identifies the sure relation SIGKDD’03, Washington DC, USA, August 2003, pp.
between these identified suspicious accounts using concepts 747-752.
like ontology. The relation between these accounts can give [5] NhienAn Le Khac, SammerMarkos, M. O'Neill, A.
us additional information like whether the involved criminal Brabazon and M-TaharKechadi. An investigation into
people are belonging to same occupation or to the same Data
location etc. The frequent accounts should not be the only Mining approaches for Anti Money Laundering. In
criteria for finding out the suspicious transaction as there may International conference on Computer Engineering &
be a case when the transaction does not occur frequently but Applications 2009.
even then they are illegal. To trace out such cases additional [6] Nhien An Le Khac, M.Teharkechadi. Application of
parameters have to be considered. Data mining for Anti-money Detection: A case study.
IEEE International conference on Data mining workshops
2010.
Acknowledgment [7] Nhien An Le Khac, SammerMarkos,M.Teharkechadi,.
A data mining based solution for detecting suspicious
I acknowledge the management, HOD, faculties and of money laundering cases in an investment bank. IEEE
coordinator computer science & engineering department for Computer society 2010.
providing the Research assistantship to carry out the above [8] J.Han and M. Kamber, Data Mining: Concepts and
work. Techniques. Morgan Kaufmann publishers, 2nd Eds., Nov
2005
References [9] Senator, T. E., Goldberg, H. G., Wooton, J., Cottini, M.
A., Khan, A. F. U., Klinger, C. D., Llamas, W. M.,
[1] R. C. Watkins et al, Exploring Data Mining technologies Marrone, M. P., Wong, R. W. H., 1995. Financial Crimes
as Tool to Investigate Money Laundering. Journal of Enforcement Network AI System (FAIS) Identifying
Policing Practice and Research: An International Journal. Potential Money Laundering from Reports of Large Cash
Vol. 4, No. 2, January 2003, pp. 163-178. Transactions. AI magazine.

8
 [16] Zengan Gao and Mao Ye. A framework for data
[10] Liu Keyan, Yu Tingting. An improved Support vector mining-based anti-money laundering research
Network Model for Anti-Money Laundering [17] R.Chandni, G. Deepa Sakthi, B.Geerthana,
[11] Assem Khalaf Ahmed, Allam El-Din, Nashaat El A.Rajalakshmi. A Graph Based Approach to Identify the
Khamesy. Data Mining Techniques for Anti Money Suspicious Accounts by Implementing the Hash Based
Laundering Association Mining
[12] Mahesh Kharote, V. P. Kshirsagar. Data Mining Model [18] Ch. Suresh, Dr. K. Thammi Reddy, N. Sweta. A
for Money Laundering Detection in Financial Domain Hybrid Approach for Detecting Suspicious Accounts in
[13] Andrea Fronzetti Colladon, Elisa Remondi. Using Money Laundering Using Data Mining Techniques
social network analysis to prevent money undering [19] Nhien-An Le-Khac, Sammer Markos and M-Tahar
[14] Z. Zang, J.J. Salermo,P. S. Yu. Applying Data mining Kechadi A Heuristics Approach for Fast Detecting
in Investigating Money Laundering Crimes Suspicious Money Laundering Cases in an Investment
[15] Ahmad Salehi, Mehdi Ghazanfari, Mohammed Bank
Fathian. Data Mining Techniques for Anti Money [20] Alexandre, Claudio Balsa, Jo˜ao Client Profiling for
Laundering an Anti-Money Laundering System

View publication stats