H. SCORE FUSION OF FINGER VEIN AND FACE
FOR HUMAN RECOGNITION BASED ON CONVOLUTIONAL NEURAL NETWORK
MODEL. International Journal of Computing10.1109/IDAACS.2017.8095208Krysovatyy, Hrystyna Lipyanina-Goncharenko, Svitlana Sachenko and OksanaWest Ukrainian National University
,
Lvivska Str., 11, Ternopil, 46000
,
Ukraine201968800000002
Fictitious business - is the creation or acquisition of business entities in order to cover up illegal activities or activities that are prohibited. Investigation of economic crime takes a lot of time for law enforcement officers, so in this regard, the development of an algorithm for detecting a fictitious enterprise based on the classic method of machine learning, namely Support Vector Machine Classification, will develop a single software environment for rapid detection of economic crimes. To build the method, data from 1,100 companies operating in Ukraine were used. The data presented in in the set logical binary values are from 355 fictitious enterprises. Modeling of the Support Vector Machine was performed by 3 approaches: linear, polynomial and radial. The best results are obtained from classification by polynomial approach. The training sample showed evaluation results at 100%, and the test sample showed evaluation at 99.7%. Also, the confusion matrix showed quite good results. Machine Classification. fictitious enterprises, business entities, classification, machine learning, Support Vector MoMLeT+DS 2021: 3rd International Workshop on Modern Machine Learning Technologies and Data Science, June 5, 2021, Lviv-Shatsk,
DesyatnyukAndriy Krysovatyy, Hrystyna Lipyanina-Goncharenko, Svitlana Sachenko and Oksana
[
1
].
crimes.
Ukraine
1. IntroductionA fictitious enterprise in Ukraine should be understood the following: a business entity that is
registered in violation of the established procedure (legal norms) of registration with state bodies, the
constituent documents of which do not comply with applicable law, or to carry out activities contrary
to law or constituent documents, or violation of the procedure for tax accounting and deadlines for filing
tax returns and financial statements, or violation of the deadlines for submission of information to
government agencies about the change of name, organizational form, form of ownership and location
The main reasons for the emergence and existence of economic crime and fictitious entrepreneurship are: imperfection of legislation governing economic activity, high levels of corruption, bondage of taxes, control of corrupt individuals in major industries, low professional level of law enforcement officers in detecting, documenting, investigating these crimes.However, the investigation of economic crime often takes a lot of time for law enforcement officers, so in this regard, the development of an algorithm for detecting a fictitious enterprise based on the classical method of machine learning, namely Support Vector Machine Classification, will develop a single software environment that is one of the most promising areas for the rapid detection of economicThis article is devoted to this topic, the rest of which is distributed as follows. Section 2 discusses the analysis of related work; section 3 presents the algorithm for detecting a fictitious enterprise based
2021 Copyright for this paper by its authors.
on Support Vector Machine Classification; in section 4 the implementation of the algorithm itself.
Section 5 presents the conclusions of the study.2. Related work
Paper [
3
] forms the basis for the analysis of the interaction of legal and illegal entities: the definition
of the terms “international crime” and “cross-border crime”; separates for analytical purposes
“enterprise crime” from “political crime”. Organized crime focuses on the “model” of the enterprise,
which is really [
4
] an approach to the study of organized crime, based on the idea that legal and illegal
business are quite similar. Based on the paradigm of organized crime, paper [
5
] presents the concept of
criminal entrepreneurship, the work developed a research model that links the degree of
entrepreneurship with the stage of organized crime growth. The article [
6
] has the application of known
theories of organized crime, such as agency theory, alliance theory, network theory, resource-based
theory and other organizational theories to shed light on criminal organizations. Theoretical and applied
aspects of the impact of criminal activity on the economic security of business structures have been
studied [
7
]; the structural and dynamic tendency of the main threats to the economic security of
enterprises on the basis of the application of integrated assessment depending on partial indicators is
studied; the peculiarities of the multi-vector mechanism of counteracting crime with the help of
economic security of enterprises are substantiated as the base of measures set choosing.
The study [
8
] considered the measure of corporate tax evasion, which reduces both financial and
taxable income, which is called tax evasion. Simulation analyzes, transformations of US LIFO / FIFO
inventory methods, and samples from private and public firms are used to confirm the results. The
results indicate that the measure of compliance with tax avoidance successfully records transactions
that comply with the accounting tax. As expected, it was found that the degree of participation of
stateowned firms in compliance with tax evasion systematically changes depending on the pressure on the
capital market.
It has been studied [2] as illegal mining, which is very common in Colombia, to overcome the
problem of measuring illegal activity, a new data set has been built using machine learning methods on
the satellite images functions. The study [
9
] listed opportunities in the virtual environment through ICT,
such as FinTrack software, etc. to prevent financial crime. Face recognition plays a central role in many
security programs [
10
], which are used to establish a huge system of social credit to control the entire
population.
A generalized structure of a high-performance adaptive system for detecting cyberattacks based on
neural networks and artificial immune systems has been developed [
15, 28
]. Various data from machine
learning technologies applied to crime data to monitor the impact of the economic crisis on crime in
India have been investigated in [
16
]. Possibilities of detailed study of data from huge repositories,
analysis of various socio-economic factors related to crime incidents, detection of deviations,
classification of patterns and development of effective computational models for crime prediction using
data analysis and machine learning have been studied [17]. The proposed [18] system is tested for the
problem of predicting crimes using data, and experimental results show that the proposed system
provides better results and search for possible solutions and patterns of crime. The experiment [19-21]
collected data on the crime scenario from the police, which were then simulated on the data set using
machine learning algorithms to predict some attributes.
The researches [11, 23, 24] present a modern, structured and well-organized review of one-class
support classifiers. The study [
12
] used the methods of vector support machines (SVM) and neural
networks, where the SVM model received the highest efficiency among the classifiers for each data set.
Research [
13
] provides an understanding of the use of stochastic gradient descent algorithms for large
data applications, for example, to accelerate SVM or controlled regression on a large scale, or to
increase the effectiveness of online learning or real-time forecasting (control). [22, 29] proposed an
approach to machine learning for crime detection and crime location, through Twitter posts and
vectorbased filtering to eliminate noise.
It should be noted that the above-mentioned works do not describe the detection of fictitious
enterprises with the help of information technology. Accordingly, the aim of this article is to develop a
method of detecting a fictitious enterprise based on Support Vector Machine Classification, which will
further develop a software environment for public sector employees to prevent economic crimes and
quickly track fictitious enterprises.
There are several close analogues to the purpose of the study [5, 6, 17], which analyze the applied area, but they do not investigate the use of information technology in this matter and further software development.3. Materials and methodsThe above-described factors in the study lead to mass shadowing of the Ukrainian economy [30].
Accordingly, there is an urgent need to develop special state programs to combat crime in the economic
sphere (public procurement, use of state budget funds, taxation of economic entities, works, services at
the expense of the state budget, etc.), certain areas of the national economy.
In order to properly organize the activities of law enforcement agencies to detect, promptly verify
and investigate crimes committed using the capabilities of economic entities with signs of fictitiousness,
it is necessary to determine their characteristics and features of the content of criminal activity and types
of such entities. In such case the developed method may be used (Fig. 1.) to identify a fictitious
enterprise on the basis of Support Vector Machine Classification, which will allow to quickly make
decisions about the performance of an individual enterprise.
Request to identify a fictitious enterprise 1
User data entry: ID,
At the first stage (block 1) the user needs to submit a request to identify a fictitious enterprise.
To build a method based on Support Vector Machine Classification, it is first needed to obtain a set
of data on which the Support Vector Machine Classification algorithm will be formed. In future research
it is planned to develop ready-made software. Accordingly, the required data to be entered (block 2)
directly by the user into the system (Table 1): company code (generated automatically in the system),
legal address, physical address, codes of economic activities, names of managers (maybe several),
photos of equipment with geolocation. All these parameters can be supplemented with new ones and
generalize the existing ones, the algorithm is easily adapted.
Based on the data entered by the user, the analysis of the following parameters (Table 2) (block 3) and search for appropriate values from sources of information created using the appropriate API and photo processing method of pattern recognition, which is the purpose of further research.
bool
bool
bool
Data type
int64
object
object
object
object
object
object.jpg
Company codeDetermining the fictitiousness of the enterprise.Availability of a register of legal entities and individuals in a single databaseAvailability of VAT, SSC and a single tax in the databaseCarrying out timely payment of taxes PO P K
L
VKK
K205
ZMI
ZD
TovZ SPAvailability of settlements with co-agentsInformation on the presence of company executives bool in the state register of declarationsAvailability of licenses in accordance with the NACE boolThe presence of criminal cases under Art. 205 of theCriminal Code of UkrainePresence of mentions of company executives with keywords: criminal case, corruption, offshore accounts, etc.Availability of land at the legal or physical address bool bool boolAvailability of registered trademarks and services, bool
database of industrial marks, database of inventions
and other databases of the Institute of Industrial
Property of UkraineAvailability of issued motor third party insurance bool policies, MTIBU policy check, motor third party database, search by state car number, check of the
Parameter
A
A&SP
E
F
AR
ZR
KR
LB
NM
NS
FR
FC
Explanation
Data
type
status of the Green Card policy for cars owned by the
company
The presence of cars and their owners issued to the bool company.Coincidence of registered cars with insurance policies boolAvailability in the database of exporters boolAvailability in the stock market database boolThe presence of cars and their owners registered for bool the company is wantedThe presence of weapons of the owners of the bool company is wantedThe presence of cultural values of the owners of the bool company is wantedAvailability of building licenses in the company boolThe presence of real estate in the company boolAvailability of the company's website boolAvailability of equipment, recognition of equipment bool by the available photo and determination of compliance of geolocation with the production addressAvailability of the company's social networks and bool affiliates
Photo of equipment with geolocationFacebook.comNext, a comparison of data (block 4) with each other, namely: FR with Foto, to determine whether the geolocation coincides with one of the addresses of the company and whether the photo shows the relevant equipment; SP with A to check whether the registered car companies match the insurance policies.Then all the data are transferred to the storage database (block 5) after which the data is converted into binary values (block 6).Add the fit parameter (block 7), which shows the value relative to whether the company is fictitious or not. Of course, machine learning algorithms operate on numerical values, so we assign the corresponding discrete values 0 or 1.It is needed always display the basic statistical characteristics of each numerical feature to verify
that all data are displayed correctly. Accordingly, we display (block 8): the number of values, the
average value, the minimal and maximal values. The std line shows the standard deviation (which
measures how scattered the values are). 25%, 50% and 75% of the rows show the corresponding
percentages of values in the corresponding parameter. With the correct visualization of the data, it is
clear trends and patterns, the ratio of variables, which allows very well notice the trends in the figures.
Therefore, graphs of data density are also displayed.Before data investigation based on machine learning, one of the most important steps should be to distribute the data (block 9). The data are divided into two groups: training set 80%, test set 20%. The training kit should be used to build machine learning models. The test kit (block 10) should be used to see how well the model works on unfamiliar data.Next, we classify the data using the method of the classical method of machine learning Support
Vector Machine Classification (block 11) [25, 26]. For the purpose of fictitious enterprises selection, it
is necessary to solve the problem of binary classification. Initially, the algorithm is trained on objects
(block 10) from the training sample, for which the class labels are known in advance. Next, the already
learned algorithm predicts the class label for each object from the deferred test sample. Class labels can
take the values = {−1, +1}.Object – is a vector with N signs х= (х1, х2, … , х ) in the n space.
When learning, the algorithm must construct a simple function ( ) = , where the argument is an
object from the n space and gives a label of class i. The main goal of Support Vector Machine (SVM)
as a classifier is to find the equation that divides the hyperplane [
14
] 1 1 + 2 2 + ⋯ + +
0 = 0 in n space, which would divide the two classes optimally. General view of the transformation
of an
object
into
a label of
class : ( ) =
( 1, 2, … , ), = − 0. After adjusting the weights of the algorithm
(
− ).
And
denote
=
and w and b, all objects
that fall on one side of the constructed hyperplane will be defined as the first class, and objects that fall
on the other side – the second class. Inside the function
() is a linear combination of object features
with algorithm weights, that is why SVM refers to linear algorithms. The hyperplane partition can be
constructed in different ways, but in SVM the weights
and are adjusted so that the class objects lie
as far as possible from the hyperplane distribution. In other words, the algorithm maximizes the gap
(Margin) between the hyperplane and the objects of the classes that are closest to it (Fig. 2). Such objects
are called reference vectors. If we consider the positive and negative sides of SVM, the positive thing
is that: SVM
works well with a space of features of large size, with data of small volume; SVM
maximizes the separation of the hyperplane and objects, which reduces the number of classification
errors; also, the algorithm is reduced to solving the problem of quadratic programming in the 3D
domain, which makes it possible to divide the hyperplane with certain hyperparameters of the
algorithm. The disadvantages of this method include the fact that: it takes a long time to learn, especially
for large data sets; instability to noise.
Support vector
learning algorithms is to ensure that each algorithm is evaluated equally on the same data.
4. Experimental results and discussionPython was chosen as programming language, as this programming language works best with machine-based data analysis. The following libraries were used for analysis: pandas, numpy, train_test_split, SVC and cross_val_score.To build the method, data from 1,100 companies operating in Ukraine were used. The data are presented in logical binary values (Fig. 3.). There are 355 fictitious enterprises in the set (Fig. 4).The degree of dependence between the indicators with the definition of a fictitious enterprise is
presented in the form of a correlation matrix (Fig. 5). The matrix shows that the fit indicator has a low
correlation with other parameters, so it is difficult to clearly determine the dependence on individual
parameters of whether the company is conducting economic crime or not. Therefore, it is advisable to
use machine learning algorithms that will more accurately determine economic crime. To do this,
modeling using the method of classical machine learning Support Vector Machine.
Modeling of the Support Vector Machine was performed by 3 approaches: linear (Fig. 6),
polynomial (Fig. 7) and radial (Fig. 8). Other approaches also exist, but are used less frequent.
We will first consider the linear approach (Fig. 6), the evaluation of the simulation was performed
on a test sample and training. The training sample showed evaluation results at 98.05%, and the test
sample showed evaluation at 97.88%. The standard deviation is 98% for both samples.
An important result is the Confusion Matrix. The different values of the Confusion matrix will be as
follows for the training sample:
• True positive (TP) = 143; this means that 143 indicators of positive class data are correctly
classified by the model;
• True negative (TN) = 612; this means that 612 data points of negative class were correctly
classified by the model;
• False positive (FP) = 15; this means that 15 indicators of negative class data were incorrectly
classified as models belonging to the positive class;
• False negative (FN) = 0; this means that 0 data indicators of the positive class were incorrectly
classified as models belonging to the negative class.
For the test calibration, the confusion matrix shows that 158 and 268 indicators of correctly defined positive and negative class, respectively, 7 indicators of incorrectly defined negative class and 0 indicators of data of positive class were incorrect.Next, consider the polynomial approach (Fig. 7). The training sample showed evaluation results at
100%, and the test sample showed evaluation at 99.7%. The standard deviation is close to 100% for
both samples. For the training calibration, the confusion matrix shows that 158 and 612 indicators of
correctly defined positive and negative class, respectively, and 0 indicators are incorrectly defined. In
the test sample, the confusion matrix showed that 61 and 268 indicators were defined correctly in the
positive and negative classes, respectively, and 1 indicator was determined incorrectly.
The last approach is radial (Fig. 8), which showed not very good results. The training sample showed
evaluation results at 79.8%, and the test sample showed evaluation at 81.21%. The standard deviation
is close to 80% for both samples. For the training calibration, the confusion matrix shows that 158
indicators are incorrectly defined, 612 indicators are correctly defined and 0 indicators are defined
correctly and incorrectly. In the test sample, the confusion matrix showed that 268 indicators were
defined correctly, but 62 indicators were determined incorrectly.
When considering 3 approaches of the Support Vector Machine, the best results are obtained when
classifying by the polynomial approach. The training sample showed evaluation results at 100%, and
the test sample showed evaluation at 99.7%. Also, the confusion matrix showed quite good results.
Thus, returning to the analogues discussed above [5, 6, 17], the developed method makes it possible to develop a single software environment for public sector employees to prevent economic crimes and quickly track fictitious enterprises.5. ConclusionsA method of detecting a fictitious enterprise based on the Support Vector Machine is proposed,
which allows to quickly track fictitious enterprises, which is useful for public sector employees to
prevent economic crimes.
The developed method is implemented on the basis of data of 1100 companies that conducted
economic activity in Ukraine and 355 of which were defined as fictitious. The results of experimental
studies showed that when considering 3 approaches of the Support Vector Machine, the best results
were obtained when classifying the polynomial approach, where the training sample showed evaluation
results at 100%, and the test sample showed evaluation at 99.7%. The confusion matrix for the training
set shows that 158 and 612 indicators of correctly defined positive and negative class, respectively, and
0 indicators are incorrectly defined. In the test sample, the confusion matrix showed that 61 and 268
indicators were defined correctly in the positive and negative classes, respectively, and 1 indicator was
determined incorrectly.
In further scientific research it is planned to conduct a more detailed analysis of the detection of
fictitious enterprises based on the methods: Gaussian Naive Bayes, Logistic Regression.
6. References
[1]
Yu
. Piliukov,
Ю.Fictitious entrepreneurship in Ukraine. The concept and connection with other economic crimes
.
Actual problems of jurisprudence
,
1
(
2019
):
140
-
144
. (in Ukrainian)
[2]
S.Saavedra
, &
M. Romero
,
Local incentives and national tax evasion: The response of illegal mining to a tax reform in Colombia
, SIEPR Discussion Paper No.
17
-
009
, Stanford Institute for Economic Policy Research, (
2017
). https://siepr.stanford.edu/sites/default/files/publications/17- 009_0.pdf.
[3]
N.Passas
,
Cross-border crime and the interface between legal and illegal actors
,
Security Journal16
(
2003
)
19
-
37
. doi:
10
.1057/palgrave.sj.
8340123
.
[4]
D.Liddick
,
The enterprise “model” of organized crime: Assessing theoretical propositions
,
Justice Quarterly16
(
1999
)
403
-
430
. doi:
10
.1080/07418829900094191.
[5]
P.Gottschalk
,
Illegal entrepreneurialism as determinant of organised business crime maturity
,
International Journal of Business and Systems Research3
(
2009
)
297
. doi:
10
.1504/ijbsr.
2009
.
026185
.
[6]
P.Gottschalk
,
How criminal organisations work: Some theoretical perspectives
,
The Police Journal: Theory, Practice and Principles81
(
2008
)
46
-
61
. doi:
10
.1350/pojo.
2008
.
81
.1.400.
[7]
O.Vivchar
,
The influence of crime activity on economic security of enterprise structures in postconflict conditions: identification of threats and mechanisms of antiaction
,
Actual Problems of Law
,
1
(
2019
)
113
-
119
. (in Ukrainian) doi: 10.35774/app2019.
01
.113.
[8]
B.Badertscher
,
S.Katz
,
S. O.Rego
, & R. J. Wilson,
Conforming tax avoidance and capital market pressure
,
The Accounting Review94
(
2019
)
1
-
30
. doi:
10
.2308/accr-52359.
[9]
K.Olowu
, & P. Gabasa, Financial crime, ICT & E-governance:
Libraries role
,
Advances in Social Sciences Research Journal7
(
2020
)
609
-
612
. https://doi.org/10.14738/assrj.71.7476.
[10]
R.T.Kreutzer
,
M.Sirrenberg
,
Fields of application of artificial intelligence - Security sector and military sector
,
in: Understanding Artificial Intelligence. Management for Professionals
, Springer, Cham,
2020
, pp.
225
-
233
. https://doi.org/10.1007/978-3-
030
-25271-
7
_
9
.
[11]
S.Alam
,
S. K.Sonbhadra
,
S.Agarwal
, & P. Nagabhushan,
One-class support vector classifiers: A survey, Knowledge-Based Systems
, (
2020
)
105754
. doi:
10
.1016/j.knosys.
2020
.
105754
.
[12]
N. L.Costa
,
L. A. G.Llobodanin
,
I. A.Castro
, &
R. Barbosa
,
Using support vector machines and neural networks to classify Merlot wines from South America
,
Information Processing in Agriculture 6
(
2018
)
265
-
278
. doi:
10
.1016/j.inpa.
2018
.
10
.003.
[13]
W.He
, & Y. Liu,
To regularize or not: Revisiting SGD with simple algorithms and experimental studies
,
Expert Systems with Applications112
(
2018
)
1
-
14
. doi:
10
.1016/j.eswa.
2018
.
06
.026.
[14]
A.Blumer
,
A.Ehrenfeucht
,
D.Haussler
, &
M. K. Warmuth
,
Learnability and the VapnikChervonenkis dimension
,
Journal of the ACM36
(
1989
)
929
-
965
. doi:
10
.1145/76359.76371.
[15]
Komar
,
M.
,
Golovko
,
V.
,
Sachenko
,
A.
,
Bezobrazov
,
S.Intelligent system for detection of networking intrusion
.
Proceedings of the 6th IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications
, IDAACS'
2011
, pp.
374
-
377
[16]
Komar
,
M.
,
Kochan
,
V.
,
Dubchak
,
L.
,
Sachenko
,
A.
,
Golovko
,
V.
,
Bezobrazov
,
S.
,
Romanets
,
I.High performance adaptive system for cyber attacks detection
,
in: Proceedings of the 2017 9th