Next Article in Journal
Practical Applications of Diffusive Realization of Fractional Integrator with SoftFrac
Previous Article in Journal
Compact and Simple High-Efficient Dual-Band RF-DC Rectifier for Wireless Electromagnetic Energy Harvesting
 
 
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Blockchain-Enabled Transaction Scanning Method for Money Laundering Detection

1
Faculty of Information Engineering, Shaoyang University, Shaoyang 422000, China
2
Department of Computer Engineering & Information Security, International Information Technology University, Almaty 050000, Kazakhstan
3
Department of Computer Science, Shaqra University, Shaqra 11961, Saudi Arabia
4
Sensor Networks and Cellular Systems Research Center, University of Tabuk, Tabuk 71491, Saudi Arabia
5
Department of Information Technology, University of Tabuk, Tabuk 71491, Saudi Arabia
*
Author to whom correspondence should be addressed.
Electronics 2021, 10(15), 1766; https://doi.org/10.3390/electronics10151766
Received: 5 June 2021 / Revised: 10 July 2021 / Accepted: 21 July 2021 / Published: 24 July 2021
(This article belongs to the Section Computer Science & Engineering)

Abstract

:
Currently, life cannot be imagined without the use of bank cards for purchases or money transfers; however, their use provides new opportunities for money launderers and terrorist organizations. This paper proposes a blockchain-enabled transaction scanning (BTS) method for the detection of anomalous actions. The BTS method specifies the rules for outlier detection and rapid movements of funds, which restrict anomalous actions in transactions. The specified rules determine the specific patterns of malicious activities in the transactions. Furthermore, the rules of the BTS method scan the transaction history and provide a list of entities that receive money suspiciously. Finally, the blockchain-enabled process is used to restrict money laundering. To validate the performance of the proposed BTS method, a Spring Boot application is built based on the Java programming language. Based on experimental results, the proposed BTS method automates the process of investigating transactions and restricts money laundering incidents.

1. Introduction

Currently, monetary transactions can be committed easily by using bank transactions. Shopping, money transfers, and ordering services are all services that are available to clients anywhere in the world [1,2]. However, these transactions open many opportunities for third parties to commit illegal activities with money without punishment [3]. One of these activities is called money laundering. Money laundering is a process whose purpose is to hide illegal sources of profit [4,5]. The need for money laundering arises in three cases. The first case is if the origin of income is criminal, e.g., illegal drug trafficking, racketeering, or corruption. Criminals receiving such income are forced to launder the money to be able to spend it freely [6,7]. The second case is if an entrepreneur or firm hides a portion of their legal income from increased taxation by underestimating revenue or overcharging, using unaccounted cash, etc. [8,9]. The third case is if the recipients of the money do not want to show their real source for security, ethical, or political reasons [10].
Anti-money laundering (AML) includes a number of measures intended to counter the legalization of proceeds of crime and curb financial flows intended for terrorist activities [11,12]. Most people who have participated in these activities [13,14] proposed methods that are based on several rules. These methods are applied to identify suspicious activities in transaction histories. During the process of reading and analyzing a significant amount of data, the human factor has an effect. Therefore, the contribution of this paper is the automation of the process by creating a Java-based application, where the rules of recognition of suspicious activity are implemented with a blockchain-enabled transaction scanning (BTS) method. AML teams suffer from false alerts that result in significant additional costs [15]. As the human resources of investigators are spent checking many false warnings, real money laundering actions continue to happen.
There are fundamental problems with existing AML systems that need to be addressed. First, they use detailed rules for each scenario, resulting in many warnings that are not actually suspicious. Second, they only check a fraction of the available data, which limits the number of signals they can use to detect money laundering. Third, they have strict data format requirements that require a painful process of data integration, which often leads to poor data quality [16,17].
The application solves these problems as follows: First, it uses only the data that the user provides. Second, it checks the user’s data in full, meticulously checking each value. Third, it allows the user to initially assign keys that will be exposed during the suspicious activity search process.
The contributions of this paper are summarized as follows:
  • Novel rules have been formulated for anomalous transaction detection that support the fight against money laundering.
  • A blockchain transaction scanning method is employed that involves the rich features of data mining and the blockchain for deciding between confirmed malicious and legitimate transactions.
For defining money laundering applications, rules such as “outlier detection” [18] or “rapid mvmt funds” are used to check transactions by comparing actions with a template. If the sequence of actions matches the template, a case can be created for the beneficiary. While other applications use difficult methods such as machine learning and hardcoding the rules into the architecture, our application will use simple methods to easily add other rules and scale the system in the future.
The remainder of the paper is organized as follows: Section 2 identifies the problem. Section 3 briefly analyzes the state-of-the-art related work. Section 4 describes the system model. Section 5 discusses the problem formulation rules for transaction anomaly detection. Section 6 defines the proposed BTS methods. Section 7 provides experimental result. Section 8 presents a conclusion of the entire paper.

2. Problem Identification

Money laundering is considered a criminal process that allows illegally earned money to flow into the basic cash flow of society. Given that financial products and services around the world fall under AML regulations, the international community believes that money laundering is a threat to the world economy. As a result of these activities, “dirty” money becomes “clean”. The point of this activity is that the origin of the illegally obtained money becomes impossible to determine, and criminals can spend it with impunity.
From a social point of view, the largest problem of money laundering is that it finances and creates favorable conditions for organized crime. Often, dirty money initially appears because of drug trafficking, tax evasion, the sale of illicit goods or trafficking, and support for terrorist acts. According to the calculations of The United Nations Office on Drugs and Crime, approximately two to five percent of global gross domestic product (GDP) (between $800 million and $2 billion) has been integrated into the global banking system through money laundering. With this in mind, it is a global problem.
This problem is so urgent that to counter it in Europe, European Union (EU) law requires companies to hire financial services to conduct checks on their clients’ AML to prevent this practice. AML measures include verifying the identity of each client by a financial service or agency and monitoring their operations. As part of the fight against money laundering, the financial institution may also request additional information from the client if it discovers any suspicious activity. The financial institution can ask the client that is depositing a large sum of money into their account to provide documents confirming the origin of the funds.
AML organizations try to define money laundering by using rules. Rules are a template of a sequence of actions that may be defined as a money laundering process. If, during an investigation, a specific beneficiary receives many alerts based on the rule matching, a case can be created for this person. The only difficulty of this entire process is that it requires a significant amount of human resources and time. Customer verification, data analysis, and mathematical calculation of revenue/waste are all the responsibility of specialists. To make the overall process easier, the most optimistic option is to automate some of these processes.

3. Related Works

The main characteristics of existing approaches are discussed in this section. The prototype application AML2ink provides a visualization of relations between accounts in transactions to further identify suspicious activity [19]. An SQL query is responsible for data processing, while GraphViz renders and produces the visualization. By using this output, the process of investigation is simplified. However, it is impossible to determine the maximum amount of data that GraphViz can render because it produces only one image file, where all entities are presented as nodes and relations as links; light data with 100 rows can require significant time to be rendered.
Kolhatkar et al. [20] fully introduced the process of a multichannel data-driven, real-time AML system, providing detailed schemas. In this approach, some methods and algorithms for defining money laundering are described. However, the paper provides little information about automatization of the entire process and no information concerning the technical aspects. Raza and Haider [21] proposed the SARDBN tool, which identifies abnormalities in the sequence of transactions. As a basic algorithm, a dynamic Bayesian network is used, which generates output by filtering transactions with outlier detection rules. Although the data of the results are diverse, they provide no useful information.
Weber et al. [22] proposed deep learning, which works with a massive amount of graph data. Then, they used graph learning to display available information to the users. Although the application provides visual data, it is impossible to manipulate the database structure. Luo [23] introduced a framework with a data mining system for detecting suspicious transactions. In this paper, specific rules such as attribute filtering and a correlation matrix between trade accounts are provided. Although the data of the results are diverse, they are useful.
Colladon and Elisa [24] proposed a social network analysis to determine money laundering. Based on network metrics, this paper presents predictive models showing the risk profiles of clients. However, this model cannot provide complete information because not everyone uses social networks. After examining these approaches, we decided that many of the proposed approaches are essential. All these approaches are aimed at improving data collection, identifying money laundering, and visualizing the outcome. Our approach will not be inferior to them; additionally, it will combine all of these advantages.

4. System Model

This section addresses the main values in transaction data for further use in rules, which are originator, beneficiary, transaction committed date, and amount of money. These four modules can be defined as follows:
  • Originator;
  • Beneficiary;
  • Transaction committed date;
  • Amount of money.
An example of bank transaction data with columns that include these modules is shown in Figure 1.
As shown in Figure 2, the transaction committed date and amount of money are marked in blue and green, respectively. The originator and beneficiary, as mentioned before, consist of multiple lines.

4.1. Originator

The account that is the source of the transaction flow. This can be described differently in the transaction history depending on the bank. For the record of the originator, the bank can use the address of the automated teller machine (ATM) from which the transaction was performed. Often, for better integration into the bank system, the address of the ATM can be divided into multiple columns. For example, the name of the country, the city, and the address can be inserted into four separate columns.

4.2. Beneficiary

The account that is the destination of the transaction flow. Its structure is the same as the originator’s. It also has columns by which it is possible to identify the location where the person received the money. It is not possible to say if the beneficiary can be the main suspect in the money laundering just because he or she receives all the money. The beneficiary can also be the originator, which means that he or she acts in the long sequence of the money laundering process.

4.3. Transaction Committed Date

The date the transaction flow occurred. The outlier detection rule identifies transaction anomalies by checking how much money the originator spends in one month, by week. The rapid mvmt funds rule uses a two-week range to check whether the originator sends or the beneficiary receives a specific percent of the money. Therefore, it is important to identify when the transaction was committed to perform an investigation.

4.4. Amount of Money

The amount of money that is used in the transaction flow. This information alone is useless. Even if the amount is very low or very high, it does not define any suspicious behavior because income is an individual element. Only by using a combination of the above elements is it possible to perform an investigation and identify anomalies in account transactions. In the money laundering process, generally, a small amount of money is used to avoid additional attention.

5. Problem Formulation Rules for Transaction Anomaly Detection

The transaction-scanning algorithm uses two rules to define anomalies in transaction data tables. The following rules are described below:
  • Outlier detection;
  • Rapid mvmt funds.

5.1. Outlier Detection

An outlier is a data point that is significantly different from the others. Conversely, inliers are data that are within a stable distribution. It is not easy to define outliers because they are highly diverse and unpredictable; however, inliers are often stable, which can help define outliers.
Outlier detection in our case works with a monthly income scenario. Most people receive a wage at the beginning of the month. Then, every week until the end of the month, the person spends a fixed amount of money. This information can be considered an inlier and can be written as follows:
T x i i = 1 n
where T is the data set of transactions with fixed expenses in one month; x is the data set of transactions for entire month; i is the sequence number of the transactions for the entire month; and n is the number of transactions within a one-month period.
Therefore, the main datum that can be useful to the rule is the expense of the last week, which can be recorded as follows:
T l x i i = 1 n
where T l is the data set of transactions with anomaly data in the last week of the month; i is the sequence number of the transactions for the last week of the month; and n is the number of transactions for the last week of the month.
Data can be defined as anomalous only if the expense of the last week is extremely high compared to the expenses of the other weeks of the month. This type of activity can be considered suspicious because most people try to save money in the last week of the month to sustain them until the next wage payment. This outlier can be defined by using the density ratio given by
W ( γ ) = p ( T ) p ( T l )
where p ( T ) is the inlier density, and p ( T l ) is the test sample density. The density ratio is close to 1 when W is an inlier, and it is close to 0 when γ is an outlier.
Theorem 1.
The highest expenses occur at the beginning of the month.
Proof. 
As a rule, at the beginning of the month, people receive their wages. Before spending this money, first, they identify their necessary costs. These required costs can be divided into the following three categories:
  • Investment or savings;
  • Mandatory payments;
  • Variable costs.

5.1.1. Investment or Savings

A certain percentage is immediately separated from the earned amount and set aside for a predetermined overall goal. Depending on the level of wealth, this percentage may vary. In most cases, people save 10% of their wages [25].

5.1.2. Mandatory Payments

After the savings funds are removed, payments that cannot be avoided will follow. First, people repay money borrowed from friends or payments on bank loans. Then, the funds required to pay for housing and communal services are calculated. Finally, the necessary costs for public transport, payment for kindergarten, medicines, gasoline for a car, etc., are deducted.

5.1.3. Variable Costs

This includes all other family expenses, such as food, shoes and clothing, household expenses, spouses’ personal expenses, entertainment, holidays, birthdays, vacations, and unexpected expenses.
The formula of the weekly expense can be written as
E w n = E i n v + E m a n d + E c o s t s
where n identifies the week number ( 1 n 4 ) , and E i n v , E m a n d , and  E c o s t s are expenses for investment, mandatory, and variable costs, respectively. As mentioned before, only in the first week, after receiving wages, do people primarily pay investment and mandatory payments because they are necessary. Thus, for the next three weeks they are not considered, as follows:
E w 2 , 3 , 4 = E c o s t s
Corollary 1.
As a result, people spend more money in the beginning of the month because of expenses for needs and budget allocation. The outlier detection process is described in Figure 3.
The outlier detection process is explained in Algorithm 1.
Algorithm 1: Outlier Detection
  • Input: T [ ] , T l , T s in
  • Output: S out
     1:
    Initialization: { T [ ] : entity’s list of weekly transaction expenses; T l : entity’s last week transaction expense; T s : size of T; S: entity’s suspicious state; M: median of transaction expenses }
     2:
    if T s is odd then
     3:
    Set M T [ T s 2 ]
     4:
    else
     5:
     Set M T T s 2 + T T s 2 + 1 2
     6:
    end if
     7:
    if M × 2 T l then
     8:
    Set S True
     9:
    else
     10:
    Set S False
     11:
    end if
Algorithm 1 explains the detecting anomaly transaction process by using outlier detection. At the beginning of the algorithm, the input and output are shown, respectively. In step 1, the initialization process of given variables is explained. Steps 2 through 6 check the size of transactions for odd/even and according to their set median transaction expense. Steps 7 through 11 check the median at the expense of the last week. If the expense exceeds the median, then the entity’s state is set to suspicious; otherwise, it is not.

5.2. Rapid Mvmt Funds

Most people spend their money by buying something or lending to someone. The rapid mvmt funds rule, for the most part, is related to the second scenario. This rule is based on transferring money from one account to another. To launder money, scammers divide it and send it to multiple accounts. These accounts can also repeat this action. As a result, many accounts will have a portion of the money; then, a reverse process will begin in which accounts will collect the money by sending it to one account. In the end, the last beneficiary will receive the laundered money.
The rapid mvmt funds process uses two weeks of income and outflow of a specific entity. The main rule of this method is that if income is between 80 and 120% of outflow during two weeks’ transactions, then the entity can be suspected to be an actor of money laundering. This can be written as follows:
S = l = 0 n t T o
S = l = 0 n t T b
where n t is the number of transactions over a two-week period, and T o and T b are transactions where the entity acted as an originator and beneficiary, respectively.
Total inequality can be written as follows:
S × 0.8 R S × 1.2
Theorem 2.
Defining money laundering by inspecting remittance transactions is difficult.
Proof. 
The money transfer industry is currently growing rapidly. In 2018, more than $689 billion in money transfer transactions were made. Hence, we can conclude that this is a good platform for money laundering [26].    □
Key risks associated with remittances are the following:
  • Digital services: Internet money transfer services are not only more difficult for authorities to control but also allow criminals to bypass identity verification processes.
  • Prepaid cards: Every bank consumer can use prepaid cards to send and withdraw money through ATMs.
  • Third party engaging: Money launderers can hire third parties on their behalf to perform transactions. Such third parties are called money mules.
  • Ownership: Given the spread of money transfer services, to avoid the rules, money launderers can obtain ownership of a money transfer company.
  • Structuring: Different accounts may be used by money launderers to participate in multiple money transfer transactions.
The probability P m d definition of money laundering can be calculated as
P m d = N t N a × 100 %
where N t is the number of transactions, and N a is the number of accounts. The number of accounts can be converted as
N a = j = 0 m C 1 , C 2 , , C ρ 1 , C ρ j
where m is the number of unique entities, C are the columns in the transaction table that are defined as originator or beneficiary, and ρ is the number of columns of C. Thus, the number of transactions can be converted as
N t = l = 0 r d ρ
where r is the number of records in the transaction table, and d is the column “id” in the table that must be unique for every row. Thus, the entire probability of detecting the money laundering is written as
P m d = i = 0 r d ρ j = 0 m C 1 , C 2 , , C ρ 1 , C ρ j × 100 %
As observed, N a is inversely proportional to probability, which means that a large number of accounts cause a small probability of defining money laundering.
Corollary 2.
Because of the large size of the industry and common uses of money transfers in people’s lives, it is difficult to define money laundering. The rapid mvmt funds actions are depicted in Figure 4.
In Algorithm 2, the rapid mvmt funds algorithm for detecting suspicious transaction activity is explained. The input and output are shown at the beginning of the algorithm. Step 1 explains the initialization process of the given variables. Steps 2 through 4 check if the amount of sent money ranges between approximately 80 and 120% of the amount of received money; if true, then the entity’s state is set to suspicious. Steps 5 and 6 check if the amount of received money ranges between approximately 80 and 120% of the amount of sent money; if true, then the entity’s state is set to suspicious. Steps 7 and 9 set the entity’s suspicious state to false because no anomaly is detected.
Hypothesis 1.
Transaction scanning processes require little processing time because of filtered columns.
Proof. 
The abstract useful information that can be used in TS methods can be written as follows:
I e = T O e + T B e
where T O e is the entity’s transactions where he or she acted as originator, and T B e is the entity’s transactions where he or she acted as beneficiary.    □
Algorithm 2: Rapid Movements of Funds.
  • Input: T o , T b in
  • Output: S out
     1:
    Initialization: { T o : money amount of certain two-week transactions as an originator; T b : money amount of certain two-week transactions as a beneficiary; S: entity’s suspicious state}
     2:
    if T o T b × 0.8 & & T o T b × 1.2  then
     3:
    Set S True
     4:
    end if
     5:
    if T b T o × 0.8 & & T b T o × 1.2  then
     6:
    Set S True
     7:
    else if T o T b × 0.8 & & T o T b × 1.2 OR T b T o × 0.8 & & T b T o × 1.2  then
     8:
    Set S False
     9:
    end if
In the bank database, there will be a table where all transaction histories are stored. The structure of this table contains many columns because a bank system usually uses an NoSQL structured database, which means that data have no relations; thus, it is not necessary to divide data into a table. The rows of a table T r can be recorded as
T r = i = 0 R ( C r ) i
where R is the total number of rows, and  C r is the column of the row.
The transaction scanning method does not require many columns C r . It primarily uses the following process for columns:
C r = D , A , O , B
where D is the transaction committed date, A is the amount of money, O is the account of the originator, and B is the account of the beneficiary.
Therefore, the rows of the table can be rewritten as follows:
T r = i = 0 R D i , A i , O i , B i
As a result, removing data about originators and beneficiaries in transactions where entities acted as originators and beneficiaries, respectively, the above equation can be described as
I e = i = 0 n D i , A i , B i + i = 0 n D i , A i , O i
Corollary 3.
By reducing the number of rows in the table, we obtain as:
T r = i = 0 R ( C r ) i T r = i = 0 R D i , A i , O i , B i
Thus, it is possible to increase the performance of transaction scanning.

6. Proposed BTS Method for Money Laundering Detection

Money laundering poses a severe threat to financial bodies that leads to national impairment. Thus, detecting doubtful transactions concerning money laundering is of paramount significance. The BTS method has a capability to scan each transaction before processing. The BTS method is depicted in Figure 5, consisting of the following steps.
  • Content pre-processing;
  • Content mining and blockchain-enabled features.

6.1. Content Pre-Processing

The leading role of content pre-processing is to excerpt datasets from different locations. Subsequently, it merges different datasets into an integrated database, and then extraction, transformation, and loading (ETL) processes are applied as employed in [27]. This process experiences content quality issues because the financial institutions possess a different set of quality issues at the content level. Most of the problems are associated with the customer information in our case. These problems include:
  • Null or dummy values: This happens in most of the data fields of the databases except the identity of the user, the user type (individual, joint, or company), and fund name.
  • Misspelling: Usually phonetic and typo errors. Additionally, banking datasets are mostly organized in a distributed fashion to maintain security and flexibility. The heterogeneity of the contents can pose a threat to the content quality, particularly when an integrating process is required. Therefore, the basic content quality issues can be addressed using pre-processing.
Numerous datasets possess different variations between features ranging from minimum to maximum: for example, 0.001 and 10,000. If such variations occur, then there is need for scaling down to make the attribute acceptable and appropriate. This process supports various classifiers to be compatible for content processing. The scaling process for determining the new feature F n is given by
F n = F o V M i n V M a x V M i n
where F o : the original features of the contents; V M a x : maximum value of the features; and  V M i n : minimum value of the features.
It is highly important to determine all of the features from all datasets. Thus, the scaling process of identifying the new features from the entire contents is given by
F n = F o V M i n V M a x V M i n ( β V M a x )
where β V M a x : new maximum features, and β V M i n : new minimum features.
Once new feature values are obtained, then there is need for standardization of the feature value, given by
F n ( s ) = ( F o M v ) σ
Substitute M v and σ , and we obtain
F n ( s ) = F o ( 1 n i = 1 n V c i ) ( a i ) ( μ ) 2 D s i
where V c i : data content value; n: number of values; M v : mean value; σ : content standard value; D s i : dataset size; a i : each value from the data content; and  μ : data content mean value.

6.2. Content Mining and Blockchain-Enabled Features

This process collects and stores the information obtained from AML experts and uses case studies and past money laundering cases. Based on the collected and stored information, the rules for anomaly detection are generated to identify the malicious activities of the outlier and rapid mvmt funds. This process is responsible for controlling the entire data mining process by employing the rules to obtain better performance. It matches the data obtained from the warehouse with the associated rules. The rules provide three types of classifications (warning, probable, and suspicious) for each transaction to money laundering depicted in Figure 5. When the customer makes the transaction, it goes to the “Rule-matching process”, which is part of the content-mining process. The Rule-matching process consists of several rules, which are matched against each transaction. Each transaction is initially marked as a “warning transaction” and sent for further investigation. Based on the investigation, if the transaction matches more than 60% of the rules, then the transaction is considered as a “probable transaction”; if the transaction matches less than 60% of the given rules, then the transaction is determined to be a safe transaction. The transactions which match more than 60% of the given rules are sent for final investigation. If the transaction matches with the rules ≥95, then it is considered to be a “suspicious transaction”. Finally, the report of the suspicious transaction is forwarded to the blockchain-enabled feature server.
The blockchain-enabled features are stored on the blockchain-enabled server that is responsible for blocking the suspicious transactions and releasing the legitimate transactions. When the blockchain-enabled server receives a message from the content-mining process component regarding a suspicious transaction, then it declares it as a “confirmed malicious”. On the other hand, if the transaction is received as “non-suspicious”, then the blockchain-enabled server declares it as a “legitimate transaction”. Finally, the legitimate transaction is allowed for further processing. The Rule-matching process is given in Algorithm 3.
Algorithm 3: Rule-Matching Process Using Content Mining
  • Input: t in
  • Output: S t ; P t ; S u t out
     1:
    Initialization: { S t : Safe transaction; t: Transaction; P t : Probable transaction; S u t : Suspicious transaction; W t : Warning transaction; g : Rule-matching process}
     2:
    Set g
     3:
    Set W t t
     4:
    if g 60 then
     5:
    t = P t
     6:
    else
     7:
    t = S t
     8:
    end if
     9:
    if g 95 then
     10:
    t = S u t
     11:
    else
     12:
    t = S t
     13:
    end if
The probability model P r m is generated, which reports the confirmed malicious state. Let us assume that if the transaction t i is reported as “confirmed malicious” to the authorities, it has a value of 1; otherwise, its value is 0. The variables associated with the transaction i are denoted as Ψ i , which can be written as
P r m ( t i = 1 | Ψ i ) & f ( Ψ i ) 0 , 1
Minimizing the money laundering M m process is derived by
P r m t i , f ( Ψ i ) = t i log f ( Ψ i ) + ( 1 t i ) log 1 f ( Ψ i )
The content-mining process component takes the features from the iterative local search and random methods explained in [28] that help to develop the final predictive model F P r m for the confirmed malicious transaction:
F P r m ( t i ) = 1 5 k = 1 5 ( t i ) k
where ( t i ) : the matching rules which do not match with the suspicious transaction.
From the above equation, we deduce that ≥95 matching rules match with the suspicious transaction, and the remaining rules do not match with the transaction. Based on the result of the predictive model, the transaction is either considered to be confirmed malicious or a legitimate transaction determined by the blockchain-enabled server.

7. Experimental Results

To validate the quality of the BTS method, the complete model is written using the Spring Boot application based on the Java programming language. To store and retrieve data, the PostgreSQL relational database version 9.6.14 was used. Application Netdata is used for the monitoring system. The laptop configuration on which the application is executed is described in Table 1.
Mock transaction backup data provided by HSBC are used as input data. The dataset contains 21,602 rows and every row has 37 columns. The columns that are defined as originator and beneficiary are “orig_line1”, “orig_line2”, “orig_line3”, and “orig_line4” and “bene_line1”, “bene_line2”, “bene_line3”, and “bene_line4”, respectively. The column “amount” contains the amount of money in the transaction, and “TXN DATE” describes the date when the transaction was committed. The proposed BTS method is compared with the state-of-the-art methods—extracting and exploring blockchain data from eosio (Xblock-eos) [29], convolutional neural network biometric cryptosystem (CNNBC) [30], detecting cryptocurrency transactions (DCT) [31], and detection of illicit accounts (DIA) [32]—to ensure the validity of the proposed algorithm. Based on the testing, interesting results are determined for the following:
  • Outlier detection;
  • Rapid mvmt funds.

7.1. Outlier Detection

For the outlier detection method, transactions with role originators and committed dates between 1 March 2021 and 30 April 2021 are used. All these data are retrieved by using the SQL script. Further calculations are performed in the Java language, and the results are stored in the database. The outlier detection scanning process takes 138,606 ms, and as a result, 581 rows are generated. The method requires 10% of processor usage and 2 GB of RAM for processing. By using the provided results of the transaction money-laundering state, the accuracy of outlier detection is determined. As shown in Figure 6a, the method provides high accuracy because it uses Algorithms 1–3, which can define only specific actions in transactions; the chances that innocent entities could unintentionally trigger this rule are minimal. Additionally, it should be noted that the accuracy of the method is more stable than contending methods (Xblock-eos, CNNBC, DCT, and DIA), as depicted in Figure 6a. While our accuracy plateaus at approximately 98.7%, in the contending methods, the accuracy begins at 60% and grows slowly. The contending methods Xblock-eos, CNNBC, DCT, and DIA produce outlier accuracy of 83.3–92.2%. When the number of transactions increases, this greatly affects the accuracy of the contending methods depicted in Figure 6b, while the proposed BTS method remains stable and produces 99.4% outlier accuracy. Thus, it is proven that even if the number of transactions increases, this does not affect the accuracy of the proposed BTS method. On the other hand, the contending methods are not appropriate to deal with the number of increased transactions.

7.2. Rapid Mvmt Funds

For rapid mvmt funds, the same date span as the outlier detection method is used, but SQL scripts are used to retrieve transactions with originator and beneficiary roles. The rapid mvmt funds scanning process took 139,701 ms. It is required to be slightly longer than the outlier detection process because in this method, entities with originator and beneficiary roles are checked. As a result, 143 rows are generated. This process uses 2 GB of RAM and 20% of processor usage. The accuracy of rapid mvmt funds is shown in Figure 7a,b. Based on the result, it is observed that the proposed BTS and contending methods produce a lower accuracy than outlier detection. This occurs because of frequent money transfers. People often give and receive money, especially when lending and borrowing. This creates anomalies that are followed by the rapid mvmt funds rule, which triggers the algorithm. Therefore, multiple false alarms are generated. In summary, it can be observed that the application generated sufficient data. The comparison of the data with the output of the bank transaction history is depicted in Figure 6b and Figure 7a.
According to Figure 7a, the contending methods started with a 60–69% accuracy; with an increase in the transactions, the accuracy increased, reaching a maximum of 65.3–77.7%. All these processes are automated and executed within a monolith application, which provides better performance than the DCT method, where a real-time AML system is described that has no automation and a poor technical description. Although Xblock-eos and CNNBC ignore the process of creating the information, they provide visual data that are better than our method’s provided raw text. However, it seems that Weber hardcoded the process of retrieving transaction tables, which means that these methods are flexible, and for other ATMs, transaction table application editing is needed. DIA attempts to use a social network to predict the risks of money laundering; however, in reality, not every ATM stores information about originators or beneficiaries, which makes this approach useless, while our proposed BTS method uses transaction-specific columns to define originators, beneficiaries, and money laundering risks.
In Figure 7b, a comparison of our proposed method is shown with the contending methods. Data are generated for every method for comparison. Scaled with our result, CCNBC provides a good example of data generation, using SQL for processing; however, it shows the worst result because GraphViz takes a large amount of computer resources for data visualization. Our proposed BTS method provides better performance than other methods, and at the end of the process, the accuracy of the proposed BTS method is approximately 19.8 to 29.1% higher than contending methods. The main reason for obtaining better performance is the use of blockchain technology. When the number of transactions increases, this does not affect the BTS because each transaction is checked by the blockchain technology server before allowing it to proceed. The blockchain technology server only permits legitimate transactions to be processed, and malicious transactions are blocked. On the other hand, the contending approaches do not distinguish between malicious and legitimate transactions. As a result, the performance of the contending approach is decreased.
Based on the results, the performance of the BTS method and machine performance consumption are demonstrated. The overall results show that during the BTS process, the application constantly uses 2 GB of RAM, but the CPU is loaded differently according to the specific method. Outlier detection uses only transactions of an entity where he or she acts as an originator, whereas the rapid mvmt method needs both originator and beneficiary transactions to look for an anomaly. The processing requirements of the methods are as follows: outlier detection uses 10%, while rapid mvmt sometimes consumes an additional 5 to 10% more than outlier detection. Outlier detection takes 138,606 ms to process 21,602 transactions and generates 581 rows, which means that approximately every 37th transaction is suspected of involving money laundering. Additionally, the rapid mvmt method is processed five times more quickly than outlier detection because of the easy implementation of the algorithm. Table 2 shows a summary of the experimental results.
Based on the above, it is possible to say that the application can process a large amount of data and provide results in a minimum amount of time. Moreover, the provided results are more useful than the solutions proposed in [9,10,11]. The application records only the record’s generated date, the rule of a detected anomaly, and the entity that is suspected. However, this paper does not include ordinary user visualization, which is provided in [7,8,9,10] and can be the first priority. However, this can be easily fixed by including the frontend side of the application. We have elaborated the perils of automation and discussed the flaws of the BTS approach when it fails and misses the ML. The proposed BTS method inherits the shortcomings of blockchain technology because the blockchain fails to accomplish the targets due to energy issues. The miners of blockchain technology are provoked to deal with complex transactions. As a result, additional energy consumption occurs that does not make BTS ideal for the real world. The ledger is streamlined with the scanning of new transactions, which results in the energy consumption by the miners. This issue can be addressed by using consensus algorithms and permissioned networks. In the future, we will try to use the consensus algorithm and permissioned networks to avoid the failure of BTS.

8. Conclusions

This paper introduces a blockchain-enabled transaction scanning method that uses outlier detection and rapid mvmt funds rules to detect anomalies in bank transaction history. To validate the accuracy of the approach, the methods of the algorithm are executed by using a mock HSBC transaction history. Outlier detection works with one month’s income and checks if the income of the last week is suspicious. Rapid mvmt funds works with two weeks’ income and outflow and checks if income is between 80 and 120% of the outflow. Based on the simulation results, it is discovered that the outlier detection method works more accurately than rapid mvmt funds, and the algorithm in this paper does not require a super machine to execute machine learning (ML) to define the method. Outlier detection uses an algorithm to define anomaly actions that are not easy to unintentionally commit, while rapid mvmt selects transaction actions that may be innocent because of simple money transfers between accounts. Therefore, the accuracy of the second rule is low. However, a combination of these two rules provides good results that investigators could use for further cases. The proposed BTS method is compared with other methods: Xblock-eos, CNNBC, DCT, and DIA. Based on the testing process, we observed that the proposed BTS method produces better results than contending methods from the accuracy perspective. The proposed BTS method could be used by financial firms, governments, non-government organizations, and banking sectors to fight against money laundering.

Author Contributions

A.O. and A.R., conceptualization, writing, idea proposal, methodology, and results; A.T., data curation, software development, and preparation; M.A. and B.A., conceptualization, draft preparation, and visualization; C.Z. review and editing. All authors have read and agreed to this version of the manuscript.

Funding

This research received no external funding.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Haller, D.R.; Nguyen, T.; Rowney, K.T.; Berger, D.A.; Kramer, G.A. System, Method and Article of Manufacture for Managing Transactions in a High Availability System. U.S. Patent 6,363,363, 26 March 2002. [Google Scholar]
  2. Dodgson, M.; Gann, D.; Wladawsky-Berger, I.; Sultan, N.; George, G. Managing digital money. Acad. Manag. J. 2015, 58, 325–333. [Google Scholar] [CrossRef][Green Version]
  3. Adrian, T.; Mancini-Griffoli, T. The rise of digital money. Annu. Rev. Financ. Econ. 2019. Available online: file:///C:/Users/MDPI/AppData/Local/Temp/FTNEA2019001.pdf (accessed on 5 June 2021). [CrossRef][Green Version]
  4. Kruisbergen, E.W.; Leukfeldt, E.R.; Kleemans, E.R.; Roks, R.A. Money talks money laundering choices of organized crime offenders in a digital age. J. Crime Justice 2019, 42, 569–581. [Google Scholar] [CrossRef]
  5. Chong, A.; Lopez-De-Silanes, F. Money laundering and its regulation. Econ. Politics 2015, 27, 78–123. [Google Scholar] [CrossRef]
  6. Weber, J.; Kruisbergen, E.W. Criminal markets: The dark web, money laundering and counterstrategies-an overview of the 10th research conference on organized crime. Trends Organ. Crime 2019, 22, 346–356. [Google Scholar] [CrossRef]
  7. Levi, M. Money for crime and money from crime: Financing crime and laundering crime proceeds. Eur. J. Crim. Policy Res. 2015, 21, 275–297. [Google Scholar] [CrossRef][Green Version]
  8. Chaikin, D. Money laundering and tax evasion—The assisting of the banking sector. In The Handbook of Business and Corruption; Emerald Publishing Limited: Bingley, UK, 2017. [Google Scholar]
  9. Alldridge, P. Tax avoidance, tax evasion, money laundering and the problem of ‘offshore’. In Greed, Corruption, and the Modern State; Edward Elgar Publishing: Cheltenham, UK, 2015. [Google Scholar]
  10. Kurnia, D.A. Study on money laundering practices from the criminal action results of political parties. Translitera: J. Kaji. Komun. Dan Studi Media 2018, 6, 24–36. [Google Scholar] [CrossRef]
  11. Ams, J.; Kyriakos-Saad, N.; El Khoury, C.; Almeida, Y.; Robert, E.; Hagan, S. Anti-Money Laundering/Combating the Financing of Terrorism (AML/CFT). In Anti-Money Laundering/Combating the Financing of Terrorism (AML/CFT); International Monetary Fund, 2018. [Google Scholar] [CrossRef]
  12. Sullivan, K. Anti-Money Laundering in a Nutshell: Awareness and Compliance for Financial Personnel and Business Managers; Apress: New York, NY, USA, 2015. [Google Scholar]
  13. Maximillian, F.; Teichmann, J. Twelve methods of money laundering. J. Money Laund. Control. 2017, 20, 130–137. [Google Scholar]
  14. Salehi, A.; Ghazanfari, M.; Fathian, M. Data mining techniques for anti money laundering. Int. J. Appl. Eng. Res. 2017, 12, 10084–10094. [Google Scholar]
  15. Sobh, T.S. An Intelligent and Secure Framework for Anti-Money Laundering. J. Appl. Secur. Res. 2020, 15, 517–546. [Google Scholar] [CrossRef]
  16. Pol, R.F. Anti-money laundering: The world’s least effective policy experiment? together, we can fix it. Policy Des. Pract. 2020, 3, 73–94. [Google Scholar] [CrossRef][Green Version]
  17. Alldridge, P. What Went Wrong with Money Laundering Law? Springer: Berlin, Germany, 2016. [Google Scholar]
  18. Bergþórsdóttir, K. Local Explanation Methods for Isolation Forest: Explainable Outlier Detection in Anti-Money Laundering. Master’s Thesis, Delft University of Technology, Delft, The Netherlands, 28 August 2020. [Google Scholar]
  19. Singh, K.; Best, P. Anti-money laundering: Using data visualization to identify suspicious activity. Int. J. Account. Inf. Syst. 2019, 34, 100418. [Google Scholar] [CrossRef]
  20. Kolhatkar, J.S.; Fatnani, S.S.; Yao, Y.; Matsumoto, K. Multi-Channel Data Driven, Real-Time Anti-Money Laundering System for Electronic Payment Cards. U.S. Patent 8,751,399, 10 June 2014. [Google Scholar]
  21. Raza, S.; Haider, S. Suspicious activity reporting using dynamic bayesian networks. Procedia Comput. Sci. 2011, 3, 987–991. [Google Scholar] [CrossRef][Green Version]
  22. Weber, M.; Chen, J.; Suzumura, T.; Pareja, A.; Ma, T.; Kanezashi, H.; Kaler, T.; Leiserson, C.E.; Schardl, T.B. Scalable graph learning for anti-money laundering: A first look. arXiv 2018, arXiv:1812.00076. [Google Scholar]
  23. Luo, X. Suspicious transaction detection for anti-money laundering. Int. J. Secur. Appl. 2014, 8, 157–166. [Google Scholar] [CrossRef]
  24. Colladon, A.F.; Remondi, E. Using social network analysis to prevent money laundering. Expert Syst. Appl. 2017, 67, 49–58. [Google Scholar] [CrossRef]
  25. Warren, E.; Tyagi, A.W. All Your Worth: The Ultimate Lifetime Money Plan; Simon and Schuster: New York, NY, USA, 2005. [Google Scholar]
  26. FATF; MENAFATF. Money Laundering through Money Remittance and Currency Exchange Providers. The Financial Action Task Force Website. Available online: https://www.fatf-gafi.org/publications/methodsandtrends/documents/ml-through-physical-transportation-of-cash.html (accessed on 27 June 2015).
  27. Theodorou, V.; Jovanovic, P.; Abellò, A.; Nakuçi, E. Data generator for evaluating etl process quality. Inf. Syst. 2017, 63, 80–100. [Google Scholar] [CrossRef][Green Version]
  28. Jullum, M.; Løland, A.; Huseby, R.B.; Ånonsen, G.; Lorentzen, J. Detecting money laundering transactions with machine learning. J. Money Laund. Control 2020, 23, 173–186. [Google Scholar] [CrossRef]
  29. Zheng, W.; Zheng, Z.; Dai, H.-N.; Chen, X.; Zheng, P. Xblock-eos: Extracting and exploring blockchain data from eosio. Inf. Process. Manag. 2021, 58, 102477. [Google Scholar] [CrossRef]
  30. Albakri, A.; Mokbel, C. Convolutional neural network biometric cryptosystem for the protection of the blockchain’s private key. Procedia Comput. Sci. 2019, 160, 235–240. [Google Scholar] [CrossRef]
  31. Baek, H.; Oh, J.; Kim, C.Y.; Lee, K. A model for detecting cryptocurrency transactions with discernible purpose. In Proceedings of the 2019 Eleventh International Conference on Ubiquitous and Future Networks (ICUFN), Split, Croatia, 2–5 July 2019; pp. 713–717. [Google Scholar]
  32. Farrugia, S.; Ellul, J.; Azzopardi, G. Detection of illicit accounts over the ethereum blockchain. Expert Syst. Appl. 2020, 150, 113318. [Google Scholar] [CrossRef][Green Version]
Figure 1. Simple anti-money laundering process with the data mining and blockchain-enabled features.
Figure 1. Simple anti-money laundering process with the data mining and blockchain-enabled features.
Electronics 10 01766 g001
Figure 2. Structure of a transaction table.
Figure 2. Structure of a transaction table.
Electronics 10 01766 g002
Figure 3. Outlier detection.
Figure 3. Outlier detection.
Electronics 10 01766 g003
Figure 4. Rapid mvmt funds example.
Figure 4. Rapid mvmt funds example.
Electronics 10 01766 g004
Figure 5. Blockchain-enabled transaction scanning system.
Figure 5. Blockchain-enabled transaction scanning system.
Electronics 10 01766 g005
Figure 6. (a) The outlier detection accuracy of the proposed BTS and contending methods with the maximum 22,500 transactions; (b) The outlier detection accuracy of the proposed BTS and contending methods with the maximum 45,000 transactions.
Figure 6. (a) The outlier detection accuracy of the proposed BTS and contending methods with the maximum 22,500 transactions; (b) The outlier detection accuracy of the proposed BTS and contending methods with the maximum 45,000 transactions.
Electronics 10 01766 g006
Figure 7. (a) Accuracy for rapid mvmt funds detection of the proposed BTS and contending methods with the maximum 22,500 transactions; (b) Accuracy for rapid mvmt funds detection of the proposed BTS and contending methods with the maximum 45,000 transactions.
Figure 7. (a) Accuracy for rapid mvmt funds detection of the proposed BTS and contending methods with the maximum 22,500 transactions; (b) Accuracy for rapid mvmt funds detection of the proposed BTS and contending methods with the maximum 45,000 transactions.
Electronics 10 01766 g007
Table 1. Testing Machine characteristics.
Table 1. Testing Machine characteristics.
OSUbuntu 20.04.1 LTS
ProcessorIntel® Core™ i7-8550U CPU
@ 1.80 GHz
Processor architecturex64
Graphic cardGeForce MX150/PCIe/SSE2
Hard drive256 GB SSD
RAM15.5 GB
Table 2. Summary of experimental results.
Table 2. Summary of experimental results.
Outlier DetectionRapid Mvmt
Execution time138,606 ms139,701 ms
Number of rows generated581143
CPU usage1.40 GHz1.50 GHz
RAM usage2 GB2 GB
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Oad, A.; Razaque, A.; Tolemyssov, A.; Alotaibi, M.; Alotaibi, B.; Zhao, C. Blockchain-Enabled Transaction Scanning Method for Money Laundering Detection. Electronics 2021, 10, 1766. https://doi.org/10.3390/electronics10151766

AMA Style

Oad A, Razaque A, Tolemyssov A, Alotaibi M, Alotaibi B, Zhao C. Blockchain-Enabled Transaction Scanning Method for Money Laundering Detection. Electronics. 2021; 10(15):1766. https://doi.org/10.3390/electronics10151766

Chicago/Turabian Style

Oad, Ammar, Abdul Razaque, Askar Tolemyssov, Munif Alotaibi, Bandar Alotaibi, and Chenglin Zhao. 2021. "Blockchain-Enabled Transaction Scanning Method for Money Laundering Detection" Electronics 10, no. 15: 1766. https://doi.org/10.3390/electronics10151766

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop