Content uploaded by Eugene Veniaminovich Lutsenko
Author content
All content in this area was uploaded by Eugene Veniaminovich Lutsenko on Apr 19, 2022
Content may be subject to copyright.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
1
UDC 004.8
UDC 004.8
05.00.00. Technical science
technical sciences
ASC-ANALYSIS OF THE DEPENDENCE OF
PAYMENT OF AIC EMPLOYEES ON THEIR
CHARACTERISTICS
ASC-ANALYSIS OF THE DEPENDENCE OF
PAYMENTS TO EMPLOYEES OF AIC FROM
THEIR CHARACTERISTICS
Lutsenko Evgeniy Veniaminovich
Doctor of Economics, Ph.D., professor
Scopus Author ID: 57191193316
RSCI SPIN code: 9523-7101
prof.lutsenko@gmail.com
Lutsenko Eugeny Veniaminovich
Dr.Sci.Econ., Cand.Tech.Sci., professor
Scopus Author ID: 57191193316
RSCI SPIN code: 9523-7101
prof.lutsenko@gmail.com
Martirosov Vladimir Garrievich
group student: PI1301
Kuban State Agrarian University named after I.T.
Trubilin, Krasnodar, Russia
Martirosov Vladimir Garrievich
Student groups: PI1301
Kuban State Agrarian University, Krasnodar,
Russia
The creation of artificial intelligence systems is one of
the important and promising areas for the development
of modern information technologies. Since there are
many alternatives to artificial intelligence systems, it
becomes necessary to assess the quality of
mathematical models of these systems. In this paper,
we consider the solution of the problem of identifying
classes of pay levels for company employees by their
characteristics. To achieve this goal, free access to test
source data and a technique that will help transform
this data into the form that is necessary for working in
an artificial intelligence system are required. A good
choice are databases from the
site:http://allexcel.ru/gotovye-tablitsy-excel-free. In
this work, the database was used: "Employee database
table, calculation of payments." At the same time, the
INF4 models based on the semantic measure of the
expediency of information by A. Kharkevich with the
integral criterion "Amount of knowledge" turned out
to be the most reliable in this application. The
accuracy of the model is 0.960, which is noticeably
higher than the reliability of expert estimates, which is
considered to be about 70%. To assess the reliability
of models in ASC-analysis and the "Eidos" system, the
Van Riesbergen F-criterion and its fuzzy multiclass
generalization proposed by Prof. E.V. Lutsenko are
used.
The creation of artificial intelligence systems is one
of the important and perspective directions of
development of modern information technology. As
there are many alternatives to artificial intelligence
systems, there is a need to evaluate mathematical
models of these systems. In this work, we consider a
solution of the problem of identifying classes of
levels of pay of employees on their characteristics.
To achieve this goal, it requires free access to test
the source data and methodology, which will help to
convert the data into the form needed for work in
artificial intelligence systems. A good choice is the
databases from the site: http://allexcel.ru/gotovye-
tablitsy-excel-free. In this work, we have used the
database called "The database table of employees,
payments calculation". The most reliable in this
application was the model of the INF4 based on
semantic appropriate measure of information of A.
Kharkevich with integral criteria of "Amount of
knowledge". The accuracy of the model is 0.960,
which is much higher than the reliability of expert
evaluations, which is equal to about 70%. To assess
the reliability of the models in the ACS-analysis and
the system called "Eidos" we have used F-criterion
of van Ritbergen and fuzzy multiclass generalization
proposed by Professor EV Lutsenko
Keywords: STUDY OF THE DEPENDENCE OF
PAYMENT OF AIC EMPLOYEES ON THEIR
CHARACTERISTICS
Doi: 10.21515/1990-4665-124-002
Keywords: INVESTIGATION OF THE
DEPENDENCE OF PAYMENT TO EMPLOYEES
OF AIC FROM THEIR CHARACTERISTICS
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
2
CONTENT
INTRODUCTION ................................................................................................................................................. 2
1. SYNTHESIS AND VERIFICATION OF MODELS ..................................................................................... 3
1.1 DESCRIPTION OF THE SOLUTION ................................................................................................................. 3
1.2. CONVERTING SOURCE DATA FROM HTML FORMAT TO MS EXCEL SOURCE DATA FILE ........................ 3
1.3. SYNTHESIS AND VERIFICATION OF STATISTICAL AND INTELLECTUAL MODELS ....................................... 1
1.4. TYPES OF MODELS OF THE EIDOS SYSTEM ................................................................................................. 3
1.5. MODEL VERIFICATION RESULTS ............................................................................................................... 5
2. SOLUTION OF IDENTIFICATION PROBLEMS IN THE MOST RELIABLE MODEL ............... 11
2.1. THE SOLUTION OF THE PROBLEM ....................................................................................................... 11
2.2. COGNITIVE FUNCTIONS ....................................................................................................................... 14
2.3. SWOT AND PEST MATRICES AND CHARTS ....................................................................................... 18
2.4. CLUSTER-CONSTRUCTIVE FEATURE ANALYSIS .................................................................................. 22
CONCLUSION ................................................................................................................................................... 23
LITERATURE .................................................................................................................................................... 24
INTRODUCTION
The creation of artificial intelligence systems is one of the important and
promising areas for the development of modern information technologies. Since
there are many alternatives to artificial intelligence systems, it becomes
necessary to assess the quality of mathematical models of these systems. In this
paper, we consider the solution of the problem of identifying classes of pay
levels for company employees by their characteristics.
To achieve this goal, free access to test source data and a technique that
will help transform this data into the form that is necessary for working in an
artificial intelligence system are required. A good choice is the database of test
problems for artificial intelligence systems of the UCI repository.
In this work, the database "baza dannix sotrudnikov»from the bank of
initial data on artificial intelligence tasks - the UCI repository.
To solve the problem, we use the standard features of Microsoft Office
Word and Excel, a notepad, as well as the Eidos-X++ artificial intelligence
system.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
3
1. SYNTHESIS AND VERIFICATION OF MODELS
1.1 Description of the solution
In accordance with the methodology of ASC-analysis, we will solve the
problem in four stages:
1. Converting source data from HTML format to MS Excel
intermediate files.
2. Transformation of initial data from intermediate MS Excel files into
databases of the "Eidos" system.
3. Synthesis and verification of domain models.
4. Application of models for solving problems of identification,
forecasting and research of the subject area.
1.2. Converting source data from HTML format to MS Excel source
data file
From an electronic database resourcehttp://allexcel.ru/gotovye-tablitsy-
excel-freetake a database of employees - "Employee database table, calculation
of
payments":http://allexcel.ru/images/Excel/baza%20dannix%20sotrudnikov2.xls,
in which we leave the following columns:
1. Position
2. Education
3. Floor
4. Year of birth
5. Year of employment
6. Amount of children
7. Experience
8. Department number
9. Income
Columns 1-8 are descriptive scales.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
4
Column 9 is the classification scale. This column shows the employees'
monthly income.
Training set:
Table 1 - baza dannix sotrudnikov.xls
n
Position
Education
Floor
Year
of
birth
Year of
employ
ment
Numb
er of
childr
en
Expe
rienc
e
No.
departme
nt
Incom
e
one
secretary
the average
well
1974
2001
3
thirte
en
one
51660
2
secretary
the average
well
1988
2007
0
7
one
50400
3
secretary referent
higher
well
1979
1995
2
ninet
een
one
57270
4
lawyer
higher
m
1947
1991
one
23
one
93375
5
lawyer
higher
m
1978
1997
0
17
one
93375
6
lawyer
higher
m
1988
2011
2
3
one
90000
7
accountant
higher
well
1970
1987
one
27
2
49800
eight
accountant
the average
well
1951
1995
4
ninet
een
2
49800
nine
accountant
average spec.
well
1953
2001
one
thirte
en
2
49200
ten
economist
higher
well
1962
1990
3
24
2
54780
eleve
n
economist
higher
well
1984
1996
one
eight
een
2
54780
12
Chief Accountant
higher
m
1964
2000
2
fourte
en
2
58080
thirtee
n
accountant
higher
well
1954
1994
one
20
3
64740
fourte
en
accountant
higher
well
1978
2011
0
3
3
62400
fifteen
accountant
higher
m
1949
2002
3
12
3
62920
sixtee
n
dendrologist
higher
well
1966
2007
one
7
3
66000
17
architect
higher
m
1981
1997
one
17
3
99600
eighte
en
economist
higher
m
1978
2006
3
eight
3
98400
ninete
en
economist
higher
well
1952
2001
2
thirte
en
3
100860
20
worker
the average
well
1988
2005
0
nine
4
45600
21
worker
the average
m
1982
1995
one
ninet
een
4
47310
22
worker
the average
m
1964
1993
3
21
4
47310
23
worker
the average
m
1981
1997
one
17
4
47310
24
worker
the average
m
1967
1991
3
23
4
47310
25
worker
the average
m
1983
2003
one
eleve
n
4
45980
26
worker
the average
well
1983
2000
0
fourte
en
4
46740
27
worker
the average
m
1998
2012
0
2
4
45600
28
driver
the average
m
1949
1991
2
23
4
48555
29
driver
average spec.
m
1967
2006
4
eight
4
46800
thirty
driver
the average
m
1981
2001
one
thirte
4
47190
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
5
en
31
agronomist
average spec.
well
1971
2007
0
7
4
50400
32
agronomist
average spec.
well
1954
1990
3
24
4
52290
33
engineer
higher
m
1983
2013
0
one
4
59250
34
engineer
higher
m
1992
2012
3
2
4
52800
35
engineer
higher
m
1965
1992
one
22
4
61230
36
engineer
higher
m
1981
1992
one
22
4
61230
37
laboratory assistant
the average
well
1990
2007
0
7
5
43200
38
laboratory assistant
the average
well
1991
2007
0
7
5
43200
39
laboratory assistant
the average
well
1989
2001
3
thirte
en
5
44280
40
laboratory assistant
the average
well
1982
2001
one
thirte
en
5
44280
41
technician
the average
m
1989
2006
one
eight
5
48000
42
technician
average spec.
m
1979
1995
2
ninet
een
5
56250
43
technician
average spec.
m
1993
2009
0
5
5
48000
44
engineer
higher
m
1969
1997
3
17
5
58740
45
engineer
higher
m
1968
1992
one
22
5
58740
Since the input of the initial data into the Eidos system is planned to be
carried out using its universal programming interface for importing data from
external databases, which works with MS Excel files, we will convert the data
from the html file to the xls file, for which we will perform the following
operations.
Let's copy the resulting table from MS Word to MS Excel and write it
with the name: Inp_data.xls to the folder: c:\Aidos-X\AID_DATA\Inp_data\. In
the Inp_data.xls file, add an empty column at position "A" and automatically
number all lines. As a result, we get a table of initial data, fully prepared for
processing in the Eidos system and recorded in the desired folder as a file of the
desired type with the desired name.
Automated formalization of the subject area by importing initial data from
external databases into the "Eidos" system.
To load the initial data base into the Eidos system, it is necessary to use
the universal programming interface for entering data from external tabular
databases, i.e. mode 2.3.2.2 (Figure 1).
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
6
Figure 1. Screen form of the Universal software interface for importing data into
the "Eidos" system (mode 2.3.2.2.)
In the screen form shown in Figure 5, set the settings shown in the figure:
− "Set the type of source data file Inp_data": "XLS - MS Excel-2003";
− "Specify the range of classification scale columns": "Start column of
classification scales" - 10, "End column of classification scales" - 10 (the last
column in the table);
− "Specify the range of columns of descriptive scales": "Start column of
descriptive scales" - 2, "End column of descriptive scales" - 9;
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
7
− "Setting the parameters for generating scripts or the method of
interpreting text fields": "Do not use the scenario method of ASK-analysis and
special interpretation of TXT-fields."
Then click the "OK" button. Next, a window opens where information
about the dimension of the model is placed (Figure 2). In this window, you must
click the "Exit to model creation" button.
Figure 2. Setting the dimension of the model of the "Eidos" system
Next, a window opens showing the stage of the process of importing data
from the external database "Inp_data.xls" into the "Eidos" system (Figure 3), as
well as the forecast for the completion of this process. In that window, you must
wait for the completion of the formalization of the subject area and click the
"OK" button.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
8
Figure 3. The process of importing data from the external database
"Inp_data.xls" into the "Eidos" system
As a result, classification and descriptive scales and gradations are
formed, with the use of which the initial data are encoded and presented in the
form of eventological databases. By this, the 2nd stage of the ASC-analysis
"Formalization of the subject area" is fully automated. To view the classification
scales and gradations, you must run mode 2.1 (Figure 4).
Figure 6. Classification scales and gradations
To view descriptive scales and gradations, you need to run mode 2.2
(Figure 5), and the training sample mode 2.3.1. (picture 6):
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
9
Figure 5. Descriptive scales and gradations (detail)
Figure 6. Training sample (fragment)
PARAMETERS OF SCALES AND GRADATIONS WITH ADAPTIVE LIMITS AND APPROXIMATELY EQUAL NUMBER OF OBSERVATIONS BY GRADATION
with correction of the rounding error of the number of observations over the gradation interval when moving to the
next gradation
CLASSIFICATION SCALE: code: [ 1], name: "INCOME", observance on the scale (total): 45, type/number of gradations in the scale: "Equal number of events in intervals"/4
1 Name of gradation: 1/4-{ 43200.0000000, 47310.0000000}, interval size= 4110.0000000, calculated/actual number of observations per gradation: 11/11
2 Name of gradation: 2/4-{ 47310.0000000, 50400.0000000}, interval size= 3090.0000000, calculated/actual number of observations per gradation: 11/11
3 Name of gradation: 3/4-{ 50400.0000000, 59250.0000000}, interval size= 8850.0000000, calculated/actual number of observations per gradation: 11/11
4 Name of gradation: 4/4-{ 59250.0000000, 100860.0000000}, interval size= 41610.0000000, calculated/actual number of observations per gradation: 12/12
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
DESCRIPTIONAL SCALE: code: [ 1], name: "POSITION", type/number of gradations in the scale: "Equal number of events in intervals"/14
1 Name of graduation: 1/14-agronomist
2 Name of graduation: 2/14-architect
3 Name of graduation: 3/14-accountant
4 Name gradations: 4/14-driver
5 Name of graduation: 5/14-chief accountant
6 Name of graduation: 6/14-dendrologist
7 Name of graduation: 7/14-engineer
8 Name of graduation: 8/14-laboratory assistant
9 Name of graduation: 9/14-working
10 Name of graduation: 10/14 secretary
11 Name of graduation: 11/14 secretary referent
12 Name of graduation: 12/14-technician
13 Name of graduation: 13/14-economist
14 Name of graduation: 14/14-lawyer
DESCRIPTIVE SCALE: code: [ 2], name: "EDUCATION", type/number of gradations in the scale: "Equal number of events in intervals"/3
15 Name of graduation: 1/3-higher
16 Name of gradation: 2/3-medium
17 Name of gradation: 3/3-medium spec.
DESCRIPTIONAL SCALE: code: [ 3], name: "SEX", type/number of gradations in the scale: "Equal number of events in intervals"/2
18 Name of gradation: 1/2-g
19 Name of gradation: 2/2-m
DESCRIPTIVE SCALE: code: [4], name: "YEAR OF BIRTHDAY", observance on the scale (total): 45, type/number of gradations in the scale: "Equal number of events in
intervals"/4
20 Name of gradation: 1/4-{1947.0000000, 1964.0000000}, interval size= 17.0000000, calculated/actual number of observations per gradation: 11/11
21 Name of gradation: 2/4-{1964.0000000, 1978.0000000}, interval size= 14.0000000, calculated/actual number of observations per gradation: 11/11
22 Name of gradation: 3/4-{1978.0000000, 1983.0000000}, interval size= 5.0000000, calculated/actual number of observations per gradation: 11/11
23 Name of gradation: 4/4-{1983.0000000, 1998.0000000}, interval size= 15.0000000, calculated/actual number of observations per gradation: 12/12
DESCRIPTIVE SCALE: code: [ 5], naming: "YEAR OF EMPLOYMENT", observance on the scale (total): 45, type/number of gradations in the scale: "Equal number of events in
intervals"/4
24 Name of gradation: 1/4-{1987.0000000, 1994.0000000}, interval size= 7.0000000, calculated/actual number of observations per gradation: 11/11
25 Name of gradation: 2/4-{1994.0000000, 2000.0000000}, interval size= 6.0000000, calculated/actual number of observations per gradation: 11/11
26 Name of gradation: 3/4-{2000.0000000, 2006.0000000}, interval size= 6.0000000, calculated/actual number of observations per gradation: 11/11
27 Name of gradation: 4/4-{2006.0000000, 2013.0000000}, interval size= 7.0000000, calculated/actual number of observations per gradation: 12/12
DESCRIPTIVE SCALE: code: [6], name: "NUMBER OF CHILDREN", observ.on the scale (total): 34, type/number of gradations in the scale: "Equal number of events in
intervals"/4
28 Name of gradation: 1/4-{1.0000000, 1.0000000}, interval size=0.0000000, calculated/actual number of observations per gradation: 8/8
29 Name of gradation: 2/4-{1.0000000, 1.0000000}, interval size=0.0000000, calculated/actual number of observations per gradation: 8/8
30 Name of gradation: 3/4-{1.0000000, 3.0000000}, interval size=2.0000000, calculated/actual number of observations per gradation: 9/9
31 Name of gradation: 4/4-{3.0000000, 4.0000000}, interval size=1.0000000, calculated/actual number of observations per gradation: 9/9
DESCRIPTIVE SCALE: code: [7], name: "EXPERIENCE", observ. on the scale (total): 45, type/number of gradations in the scale: "Equal number of events in intervals"/4
32 Name of gradation: 1/4-{ 1.0000000, 7.0000000}, interval size= 6.0000000, calculated/actual number of observations per gradation: 11/11
33 Name of gradation: 2/4-{ 7.0000000, 13.0000000}, interval size= 6.0000000, calculated/actual number of observations per gradation: 11/11
34 Name of gradation: 3/4-{13.0000000, 19.0000000}, interval size= 6.0000000, calculated/actual number of observations per gradation: 11/11
35 Name of gradation: 4/4-{19.0000000, 27.0000000}, interval size= 8.0000000, calculated/actual number of observations per gradation: 12/12
DESCRIPTIONAL SCALE: code: [ 8], name: "SECTION NUMBER", observance on the scale (total): 45, type/number of gradations in the scale: "Equal number of events in
intervals"/4
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
4
36 Name of gradation: 1/4-{1.0000000, 2.0000000}, interval size=1.0000000, calculated/actual number of observations per gradation: 11/11
37 Name of gradation: 2/4-{2.0000000, 4.0000000}, interval size=2.0000000, calculated/actual number of observations per gradation: 11/11
38 Name of gradation: 3/4-{4.0000000, 4.0000000}, interval size=0.0000000, calculated/actual number of observations per gradation: 11/11
39 Name of gradation: 4/4-{4.0000000, 5.0000000}, interval size=1.0000000, calculated/actual number of observations per gradation: 12/12
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
1
This creates all the necessary and sufficient prerequisites for identifying
the strength and direction of causal relationships between the values of the
factors and the results of their joint systemic impact (taking into account the
nonlinearity of the system [11]).
1.3. Synthesis and verification of statistical and intellectual models
Next, we launch mode 3.5, in which models for synthesis and verification
are specified, and a model is specified, which, at the end of the mode, is
assigned the status of the current one (Figure 7).
Figure 7. Selection of models for synthesis and verification, as well as the
current model
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
2
In this mode, there are many different methods for verifying models,
including those that support the bootstrap method. But we use the default
parameters shown in Figure 7. The stage of the 3.5 mode execution process and
the forecast of its end time are displayed on the screen form shown in Figure 8.
Figure 8. Synthesis and verification of statistical and knowledge models
It is interesting to note (see Figure 8) that the synthesis and verification of
all 10 models on this task took 18 seconds. At the same time, verification
(assessment of the reliability of models) was carried out on all 45 observation
examples from the training set. As a result of the execution of mode 3.5, all
models were created, with all the particular criteria listed in Figure 10, but
below we will present only some of them (Tables 2, 3, 4).
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
3
1.4. Types of models of the Eidos system
Consider the solution of the identification problem using the example of
the INF1 model, in which the amount of information according to A.
Kharkevich is calculated, which we receive about the belonging of an identified
object to each of the classes, if we know that this object has some attribute.
In fact, private criteria are simply formulas for converting the absolute
frequency matrix (Table 2) into matrices of conditional and unconditional
percentage distributions, and knowledge matrices (Tables 3 and 4).
Table 2 - Matrix of absolute frequencies (ABS model) and conditional and
unconditional percentage distributions (fragments)
Table 3 – Information content matrix (INF1 model) in bits (fragment)
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
4
Table 4 - Knowledge matrix (INF3 model) (fragment)
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
5
1.5. Model Verification Results
The results of verification (estimation of reliability) of models that differ
in particular criteria with the two integral criteria given above are shown in
Figure 9.
a)
b)
Figure 9. Model reliability estimates
The models INF4, INF5 turned out to be the most reliable in this
application with the integral criterion "Semantic resonance of knowledge". At
the same time, the accuracy of the model is 0.755 and the completeness of the
model is 0.822, which are quite good indicators. Thus, the level of reliability of
forecasting using the model is higher than that of expert estimates, the reliability
of which is considered to be approximately 70%. To assess the reliability of
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
6
models in ASC-analysis and the Eidos system, the Van Riesbergen F-criterion is
used, as well as its fuzzy multiclass generalization proposed by Prof. E.V.
Lutsenko [16] (Figure 10).
Figure 10. Types of forecasts and the principle of determining the reliability of
models according to the author's version of a metric similar to the F-criterion
It is also noteworthy that statistical models, as a rule, give a significantly
lower weighted average reliability of identification and non-identification than
knowledge models, and almost never a higher one. This justifies the use of
knowledge models and intellectual technologies. Figure 11 shows partial
distributions of similarity and difference levels for correctly and erroneously
identified and unidentified situations in the most reliable INF5 model.
Figure 11 shows that:
- the most reliable INF5 model better determines the non-membership of
an object to a class than membership (which can also be seen from Figure 9);
- the modulus of the similarity-difference level in the most reliable INF5
model for correctly identified and correctly unidentified objects is significantly
higher than for erroneously identified and erroneously unidentified objects. This
is true for almost the entire range of similarity-difference levels, except for small
modulo values in the range from 0 to 15% of the similarity level. For large
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
7
similarity-difference levels (more than 20%), the difference between correctly
and erroneously identified and unidentified situations is also very large.
Figure 11. Partial similarity-difference distribution of correctly and erroneously
identified and unidentified states of the simulation object in the INF5 model
This means that if we take into account not only the facts of correctly and
erroneously identified and unidentified objects, but also take into account the
level of similarity-difference, then it is possible to nullify erroneous
identifications and misidentifications and evaluate the reliability of the model
much more accurately than using the Van Riesbergen F-criterion . This idea is
the basis of the fuzzy multiclass generalization using the Van Riesbergen F-
criterion proposed by Prof. E.V. Lutsenko (L-measure) [16].
For the most reliable INF5 model, the L-score is 0.930 with the model
accuracy of 0.926 and model recall of 0.933 (see Figure 9b), which are very
good indicators.
Any observational data can be considered the sum of the true value and
noise, and neither the first nor the second is unknown. Therefore, it makes sense
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
8
to compare the created models with purely random models, coinciding in basic
characteristics. The Eidos system has laboratory work No. 2.01: "Investigation
of the RND model with different sample sizes." If this work is installed in the
absence of the current application, then all parameters of the created models are
set manually, but if the current model exists, as in our case, then all its main
parameters are determined automatically (Figure 12):
Figure 12. Screen form for managing the creation of random models that match
the current one in terms of dimensions of the main databases
Figure 13 shows the partial similarity-difference distribution of correctly
and erroneously identified and unidentified states in a random INF5 model.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
9
Figure 13. Partial distribution of similarity-difference of correctly and
erroneously identified and unidentified states in a random INF5 model
The quite obvious difference in the frequency distributions of the
similarity-difference levels of correctly and erroneously identified and
unidentified states of the modeling object and the random model (Figures 11 and
13) is explained by the fact that in real models, in addition to noise, there is also
information about the true cause-and-effect relationships of factors and their
values on the one hand, and the states of the modeling object, which are caused
by them, on the other hand. If there is no such information in the model, then the
distribution is obtained of the type shown in Figure 13.
Figure 14 shows data on the reliability of statistical and cognitive models
created on the basis of a random sample.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
10
Figure 14. Reliability of statistical and cognitive models created on the basis of a
random sample
Based on a comparison of Figures 14 and 9, the following conclusions can
be drawn:
- the reliability of the best INF5 model according to the F-criterion,
reflecting the real object of modeling, is approximately 41% higher than that of
a similar random model (0.787/0.558=1.41);
- the difference between the reliability of statistical models and
knowledge models created on the basis of a random sample is much less than
that of models that reflect the real object of modeling;
- in real models, in addition to noise, there is also information about the
true cause-and-effect relationships of factors and their values, on the one hand,
and the states of the modeling object that they cause, on the other hand, and
approximately 1/3 of the reliability is due to the reflection in real models of the
laws of the subject area , and 2/3 of the reliability is due to the presence of noise
in the original data. Based on this, it can be assumed that in the initial data, the
signal level about real cause-and-effect relationships in the simulated subject
area is approximately two times lower than the noise level.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
11
2. SOLUTION OF IDENTIFICATION PROBLEMS IN THE MOST
RELIABLE MODEL
2.1. The solution of the problem
In accordance with the ASC analysis technology, we set the current model
INF5 (mode 5.6) (Figure 15) and perform batch recognition in mode 4.2.1.
(picture 16)
Figure 15. Screen forms of the mode of setting the model as current
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
12
Figure 16. Screen form of batch recognition mode in the current INF5 model
As a result of batch recognition, a number of databases are created in the
current model, which are visualized in output screen forms that reflect the
results of solving the problem of identification and prediction.
Mode 4.1.3 of the Eidos system provides display of identification and
forecasting results in various forms:
1. In detail visually: "Object - classes".
2. In detail visually: "Class - objects".
3. Results visually: "Object - classes".
4. Results visually: "Class - objects".
5. Briefly summarized: "Object - classes".
6. Generalized form according to the reliability of models for different
integral criteria.
7. Generalized statistical analysis of the results of identification by models
and integral criteria.
8. Statistical analysis of identification results by classes, models and
integral criteria.
9. Recognition of the level of similarity under different models and
integral criteria.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
13
10. Reliability of class identification under different models and integral
criteria.
Below we will briefly review some of them.
Figures 17 and 18 show examples of forecasts in the most reliable INF5
model:
Figure 17. Example of class identification in the INF5 model
Figure 18. An example of identifying classes in the INF5 model
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
14
2.2. cognitive functions
Consider mode 4.5, which implements the ability to visualize cognitive
functions for any models and any combination of classification and descriptive
scales (Figure 19)
Figure 19. Screen form of mode 4.5 of the "Eidos-X++" system "Visualization
of cognitive functions"
With regard to the problem considered in this paper, the cognitive
function shows how much information is contained in various values of the
factors that the modeling object will go into certain future states. Many works of
the author 9 are devoted to cognitive functions, but the most recent and general
of them is the work. Therefore, we will not dwell here on the description of what
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
15
cognitive functions are in ASC analysis. Figure 20 shows the visualization of all
cognitive functions of this application for the INF5 model.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
16
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
17
Figure 20. Visualization of cognitive functions for generalized classes and all
descriptive scales in the INF5 model
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
18
2.3. SWOT and PEST matrices and charts
SWOT analysis is widely known andgenerally accepted method of
strategic planning. However, this does not prevent him from being criticized,
often quite fair, justified and well reasoned. As a result of a critical review of the
SWOT analysis, quite a few of its weaknesses (shortcomings) were identified,
the source of which is the need to involve experts, in particular, to assess the
strength and direction of the influence of factors. It is clear that experts do this in
an informal way (intuitively), based on their professional experience and
competence. But the possibilities of experts have their limitations and often for
various reasons they cannot and do not want to do this. Thus, there is a problem
of conducting a SWOT analysis without the involvement of experts. This
problem can be solved by automating the functions of experts, i.e. by measuring
the strength and direction of influence of factors directly on the basis of
empirical data. A similar technology has been developed for a long time, it is
already about 30 years old, but it is little known - this is the Eidos intellectual
system. This system has always provided the ability to conduct quantitative
automated SWOT analysis without the use of expert assessments directly on the
basis of empirical data. The results of the SWOT analysis were displayed in the
form of information portraits. In the version of the system under MS Windows:
"Eidos-X++", an automated quantitative solution of direct and inverse problems
of SWOT analysis with the construction of traditional SWOT matrices and
diagrams is proposed (Figure 21). This system has always provided the ability to
conduct quantitative automated SWOT analysis without the use of expert
assessments directly on the basis of empirical data. The results of the SWOT
analysis were displayed in the form of information portraits. In the version of the
system under MS Windows: "Eidos-X++", an automated quantitative solution of
direct and inverse problems of SWOT analysis with the construction of
traditional SWOT matrices and diagrams is proposed (Figure 21). This system
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
19
has always provided the ability to conduct quantitative automated SWOT
analysis without the use of expert assessments directly on the basis of empirical
data. The results of the SWOT analysis were displayed in the form of
information portraits. In the version of the system under MS Windows: "Eidos-
X++", an automated quantitative solution of direct and inverse problems of
SWOT analysis with the construction of traditional SWOT matrices and
diagrams is proposed (Figure 21).
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
20
Figure 21. An example of a SWOT matrix in the INF5 model
Figure 22 shows examples of an inverted SWOT matrix and an inverted
SWOT chart in the INF5 model.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
21
Figure 22. An example of a SWOT matrix in the INF5 model
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
22
2.4. Cluster-constructive feature analysis
Figure 23 shows the results of a cluster-constructive analysis of features:
Figure 23. Results of cluster-constructive analysis of features
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
23
CONCLUSION
Since there are many artificial intelligence systems, there is a need for a
comparable assessment of the quality of their mathematical models. One of the
options for solving this problem is to test various systems on a common source
data base, for which it is very convenient to use the public UCI repository
database. This paper provides a detailed example of using the UCI repository
database to assess the quality of mathematical models used in ASC analysis and
its software tools in the Eidos artificial intelligence system. At the same time,
the INF4 models based on the semantic measure of the expediency of
information by A. Kharkevich with the integral criterion "Amount of
knowledge" turned out to be the most reliable in this application. The accuracy
of the model is 0.960, which is noticeably higher than the reliability of expert
estimates, which is considered to be about 70%. To assess the reliability of
models in ASC-analysis and the "Eidos" system, Van Riesbergen's F-criterion
and its fuzzy multiclass generalization proposed by Prof. E.V. Lutsenko are
used. It is also noteworthy that the statistical models in this application give
about 21% lower weighted average reliability of identification and non-
identification than the knowledge models, which, as a rule, is also observed in
other applications. This justifies the use of knowledge models. that the statistical
models in this application give about 21% lower weighted average certainty of
identification and non-identification than knowledge models, which is generally
observed in other applications. This justifies the use of knowledge models. that
the statistical models in this application give about 21% lower weighted average
certainty of identification and non-identification than knowledge models, which
is generally observed in other applications. This justifies the use of knowledge
models.
Based on the UCI database considered in this paper, to build forecasting
models not using ASC analysis and the Eidos system that implements it, but
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
24
using other mathematical methods and software systems that implement them,
then their quality can be compared comparable.
Literature
1. Lutsenko E.V. Methodology for using the UCI repository to assess the quality of
mathematical models of artificial intelligence systems / E.V. Lutsenko // Polythematic
network electronic scientific journal of the Kuban State Agrarian University (Scientific
journal of KubGAU) [Electronic resource]. - Krasnodar: KubGAU, 2003. - No. 02 (002). pp.
120 – 145. – IDA [article ID]: 0020302012. – Access mode:
http://ej.kubagro.ru/2003/02/pdf/12.pdf, 1.625 c.p.l.
2. Lutsenko E.V. ASC-analysis, modeling and identification of living beings based on
their phenotypic traits / E.V. Lutsenko, Yu.N. Penkina // Polythematic network electronic
scientific journal of the Kuban State Agrarian University (Scientific journal of KubGAU)
[Electronic resource]. - Krasnodar: Kub-GAU, 2014. - No. 06 (100). pp. 1346 – 1395. – IDA
[article ID]: 1001406090. – Access mode: http://ej.kubagro.ru/2014/06/pdf/90.pdf, 3.125
c.p.l.
3. Lutsenko E.V. Theoretical foundations, technology and tools of automated system-
cognitive analysis and the possibility of its application for a comparable assessment of the
effectiveness of universities / E.V. Lutsenko, V.E. Korzhakov // Polythematic network
electronic scientific journal of the Kuban State Agrarian University (Scientific journal of
KubGAU) [Electronic resource]. - Krasnodar: Kub-GAU, 2013. - No. 04 (088). pp. 340 –
359. – IDA [article ID]: 0881304022. – Access mode:
http://ej.kubagro.ru/2013/04/pdf/22.pdf, 1.25 c.p.l.
4. Lutsenko E.V. Automated system-cognitive analysis in the management of active
objects (system theory of information and its application in the study of economic, socio-
psychological, technological and organizational-technical systems): Monograph (scientific
edition). - Krasnodar: KubGAU. 2002. - 605 p.
5. [Electronic resource]. Article "baza dannix sotrudnikov": http://allexcel.ru/gotovye-
tablitsy-excel-besplatno, free. - Zagl. from the screen. Yaz. eng.
6. Website of Professor E.V. Lutsenko [Electronic resource]. Access mode:
http://lc.kubagro.ru/, free. - Zagl. from the screen. Yaz. Russian
7. Lutsenko E.V. 30 years of the Eidos system - one of the oldest domestic universal
systems of artificial intelligence, widely used and developing at the present time / E.V.
Lutsenko // Polythematic network electronic scientific journal of the Kuban State Agrarian
University (Scientific journal of KubGAU) [Electronic resource]. - Krasnodar: KubGAU,
2009. - No. 10 (054). pp. 48 – 77. – Informregistr code: 0420900012\0110, IDA [article ID]:
0540910004. – Access mode: http://ej.kubagro.ru/2009/10/pdf/04.pdf, 1.875 c.p. .l
8. Lutsenko E.V. Universal cognitive analytical system "Eidos-X++" / E.V. Lutsenko //
Polythematic network electronic scientific journal of the Kuban State Agrarian University
(Scientific journal of KubGAU) [Electronic resource]. - Krasnodar: KubGAU, 2012. - No. 09
(083). pp. 328 – 356. – IDA [article ID]: 0831209025. – Access mode:
http://ej.kubagro.ru/2012/09/pdf/25.pdf, 1.812 c.p.l.
9. Lutsenko E.V. Prediction of the number and classes of solar flares based on their
prehistory according to the UCI repository using ASC analysis and the Eidos intelligent
system / E.V. Lutsenko, A.Yu. Borovko // Polythematic network electronic scientific journal
of the Kuban State Agrarian University (Scientific journal of KubGAU) [Electronic resource].
- Krasnodar: KubGAU, 2014. - No. 10 (104). pp. 1309 – 1370. – IDA [article ID]:
1041410099. – Access mode: http://ej.kubagro.ru/2014/10/pdf/99.pdf, 3.875 c.p.l.
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
25
10. Lutsenko E.V. Modeling of complex multifactorial nonlinear control objects based
on fragmented noisy empirical data of large dimensions in system-cognitive analysis and the
intellectual system "Eidos-X++" / E.V. Lutsenko, V.E. Korzhakov // Polythematic network
electronic scientific journal of the Kuban State Agrarian University (Scientific journal of
KubSAU) [Electronic resource]. - Krasnodar: KubGAU, 2013. - No. 07 (091). pp. 164 – 188.
– IDA [article ID]: 0911307012. – Access mode: http://ej.kubagro.ru/2013/07/pdf/12.pdf,
1.562 c.p.l.
11. Lutsenko E.V. Fuzzy multiclass generalization of the classical F-measure of
reliability of Van Riesbergen models in ASC-analysis and the Eidos system / Lutsenko E.V. //
Polythematic network electronic scientific journal of the Kuban State Agrarian University
(Scientific Journal of KubGAU) [Electronic resource]. - Krasnodar: KubGAU, 2016. - No. 09
(123). – Access mode: http://ej.kubagro.ru/2016/09/pdf/01.pdf, 1.813 c.u.l. – IDA [article ID]:
1231609001.http://dx.doi.org/10.21515/1990-4665-123-001
12. Lutsenko E.V. Quantitative automated SWOT- and PEST-analysis by means of
ASC-analysis and intellectual system "Eidos-X++" / E.V. Lutsenko // Polythematic network
electronic scientific journal of the Kuban State Agrarian University (Scientific journal of
KubGAU) [Electronic resource]. - Krasnodar: KubGAU, 2014. - No. 07 (101). pp. 1367 –
1409. – IDA [article ID]: 1011407090. – Access mode:
http://ej.kubagro.ru/2014/07/pdf/90.pdf, 2.688 c.p.l.
References
1. Lucenko EV Metodika ispol'zovanija repozitorija UCI dlja ocenki kachestva
matematicheskih modelej sistem iskusstvennogo intellekta / EV Lucenko // Politematicheskij
setevoj jelektronnyj nauchnyj zhurnal Kubanskogo gosudarstvennogo agrarnogo universiteta
(Nauchnyj AUnyjurnal KubG). - Krasnodar: KubGAU, 2003. - No. 02 (002). S. 120 – 145.
– IDA [article ID]: 0020302012. – Regim dostupa: http://ej.kubagro.ru/2003/02/pdf/12.pdf,
1,625 upl
2. Lucenko EV ASK-analiz, modelirovanie i identifikacija zhivyh sushhestv na osnove
ih fenotipicheskih priznakov / EV Lucenko, Ju.N. Penkina // Politematicheskij setevoj
jelektronnyj nauchnyj zhurnal Kubanskogo gosudarstvennogo agrarnogo universiteta
(Nauchnyj zhurnal KubGAU) [Jelektronnyj resurs]. - Krasnodar: Kub-GAU, 2014. - No.
06(100). S. 1346 – 1395. – IDA [article ID]: 1001406090. – Rezhim dostupa:
http://ej.kubagro.ru/2014/06/pdf/90.pdf, 3.125 upl
3.Lucenko EV Teoreticheskie osnovy, tehnologija i instrumentarij avtomati-
zirovannogo sistemno-kognitivnogo analiza i vozmozhnosti ego primenenija dlja
sopostavimoj ocenki jeffektivnosti vuzov / EV Lucenko, VE Korzhakov // Politematicheskij
setevoj jelektronnyj nauchnyj zhurnal Kubanskogo gosudarstvennogo agrarno- go universiteta
(Nauchnyj zhurnal KubGAU) [Jelektronnyj resources]. - Krasnodar: Kub-GAU, 2013. - No.
04 (088). S. 340 – 359. – IDA [article ID]: 0881304022. – Regim dostupa:
http://ej.kubagro.ru/2013/04/pdf/22.pdf, 1.25 upl
4.Lucenko EV Avtomatizirovannyj sistemno-kognitivnyj analiz v upravlenii aktivnymi
ob#ektami (sistemnaja teorija informacii i ee primenenie v issledovanii jekonomicheskih,
social'no-psihologicheskih, tehnologicheskih i organizacionno-tehnicheskih sistem):
Monografija (scientific). – Krasnodar: KubGAU. 2002. - 605 s.
5.[Jelektronnyj resources]. Stat'ja "baza dannix sotrudnikov": http://allexcel.ru/gotovye-
tablitsy-excel-besplatno, svobodnyj. - Zagl. s jekrana. Jaz. eng.
6.Site professora EVLucenko [Jelektronnyj resources]. Rezhim dostupa:
http://lc.kubagro.ru/, svobodnyj. - Zagl. s jekrana. Jaz. rus.
7.Lucenko EV 30 let sisteme «Jejdos» - odnoj iz starejshih otechestvennyh
universal'nyh sistem iskusstvennogo intellekta, shiroko primenjaemyh i razvivajushhihsja iv
Scientific journal of KubGAU, No. 124(10), 2016of the year
http://ej.kubagro.ru/2016/10/pdf/02.pdf
26
nastojashhee vremja / EV Lucenko // Politematicheskij setevoj jelektronnyj nauchnyj zhurnal
Kubanskogo gosudarstvennogo agrarnogo universiteta (Nauchnyj zhurnal KubGAU)
[Jelektronnyj resource]. - Krasnodar: KubGAU, 2009. - No. 10 (054). S. 48 - 77. - Shifr
Informregistra: 0420900012\0110, IDA [article ID]: 0540910004. - Rezhim dostupa:
http://ej.kubagro.ru/2009/10/pdf/04.pdf, 1,875 upl
8. Lucenko EV Universal'naja kognitivnaja analiticheskaja sistema "Jejdos- H++" / EV
Lucenko // Politematicheskij setevoj jelektronnyj nauchnyj zhurnal Kubanskogo
gosudarstvennogo agrarnogo universiteta (Nauchnyj zhurnal KubGAU) [Jelektronnyj resurs].
- Krasnodar: KubGAU, 2012. - No. 09 (083). S. 328 – 356. – IDA [article ID]: 0831209025. –
Regim dostupa: http://ej.kubagro.ru/2012/09/pdf/25.pdf, 1,812 upl
9.Lucenko EV Prognozirovanie kolichestva i klassov solnechnyh vspyshek na osnove ih
predystorii po dannym repozitorija UCI s primeneniem ASK-analiza i intellektual'noj sistemy
"Jejdos" / EV Lucenko, A.Ju. Borovko // Politematicheskij setevoj jelektronnyj nauchnyj
zhurnal Kubanskogo gosudarstvennogo agrarnogo universiteta (Nauchnyj zhurnal KubGAU)
[Jelektronnyj resurs]. - Krasnodar: KubGAU, 2014. - No. 10 (104). S. 1309 – 1370. – IDA
[article ID]: 1041410099. – Rezhim dostupa: http://ej.kubagro.ru/2014/10/pdf/99.pdf, 3,875
upl
10.Lucenko EV Modelirovanie slozhnyh mnogofaktornyh nelinejnyh ob # ektov
upravlenija na osnove fragmentirovannyh zashumlennyh jempiricheskih dannyh bol'shoj
razmernosti v sistemno-kognitivnom analize i intellektual'noj sisteme «Jejdos-H ++» / EV
Lucenko, VE Korzhakov // Politematicheskij setevoj jelektronnyj nauchnyj zhurnal
Kubanskogo gosudarstvennogo agrarnogo universiteta (Nauchnyj zhurnal KubGAU)
[Jelektronnyj resurs]. - Krasnodar: KubGAU, 2013. - No. 07 (091). S. 164 – 188. – IDA
[article ID]: 0911307012. – Regim dostupa: http://ej.kubagro.ru/2013/07/pdf/12.pdf, 1,562
upl
11. Lucenko E.V. - Krasnodar: KubGAU, 2016. - No. 09 (123). – Rezhim dostupa:
http://ej.kubagro.ru/2016/09/pdf/01.pdf, 1,813 upl – IDA [article ID]: 1231609001.
http://dx.doi.org/10.21515/1990-4665 -123-001
12. Lucenko EV Kolichestvennyj avtomatizirovannyj SWOT- i PEST-analiz sredstvami
ASK-analiza i intellektual'noj sistemy "Jejdos-H++" / EV Lucenko // Politematicheskij
setevoj jelektronnyj nauchnyj zhurnal Kubanskogo gosudarstvennogo agrarnogo universiteta
(NauchnyAusurelG) []. - Krasnodar: KubGAU, 2014. - No. 07 (101). S. 1367 – 1409. – IDA
[article ID]: 1011407090. – Regim dostupa: http://ej.kubagro.ru/2014/07/pdf/90.pdf, 2,688
upl