Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках
The work considers intelligent methods for solving the problem of short- and middle-term forecasting in the financial sphere. LSTM DL networks, GMDH, and hybrid GMDH-neo-fuzzy networks were studied. Neo-fuzzy neurons were chosen as nodes of the hybrid network, which allows to reduce computational co...
Збережено в:
| Дата: | 2023 |
|---|---|
| Автори: | , , |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2023
|
| Теми: | |
| Онлайн доступ: | https://journal.iasa.kpi.ua/article/view/290368 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | System research and information technologies |
| Завантажити файл: |  |
Репозитарії
System research and information technologies| _version_ | 1866302934530129920 |
|---|---|
| author | Zaychenko, Yuriy Zaichenko, Helen Kuzmenko, Oleksii |
| author_facet | Zaychenko, Yuriy Zaichenko, Helen Kuzmenko, Oleksii |
| author_sort | Zaychenko, Yuriy |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2023-11-07T22:19:24Z |
| description | The work considers intelligent methods for solving the problem of short- and middle-term forecasting in the financial sphere. LSTM DL networks, GMDH, and hybrid GMDH-neo-fuzzy networks were studied. Neo-fuzzy neurons were chosen as nodes of the hybrid network, which allows to reduce computational costs. The optimal network parameters were found. The synthesis of the optimal structure of hybrid networks was performed. Experimental studies of LSTM, GMDH, and hybrid GMDH-neo-fuzzy networks with optimal parameters for short- and middle-term forecasting have been conducted. The accuracy of the obtained experimental predictions is compared. The forecasting intervals for which the application of the researched artificial intelligence methods is the most expedient have been determined. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2023.3.04 |
| first_indexed | 2025-07-17T10:28:21Z |
| format | Article |
| fulltext |
Yu. Zaychenko, He. Zaichenko, O. Kuzmenko, 2023
54 SSN 1681–6048 System Research & Information Technologies, 2023, № 3
TIДC
ПРОБЛЕМИ ПРИЙНЯТТЯ РІШЕНЬ ТА
УПРАВЛІННЯ В ЕКОНОМІЧНИХ, ТЕХНІЧНИХ,
ЕКОЛОГІЧНИХ І СОЦІАЛЬНИХ СИСТЕМАХ
UDC 519.925.51
DOI: 10.20535/SRIT.2308-8893.2023.3.04
INVESTIGATION OF COMPUTATIONAL INTELLIGENCE
METHODS IN FORECASTING AT FINANCIAL MARKETS
Yu. ZAYCHENKO, He. ZAICHENKO, O. KUZMENKO
Abstract. The work considers intelligent methods for solving the problem of short-
and middle-term forecasting in the financial sphere. LSTM DL networks, GMDH,
and hybrid GMDH-neo-fuzzy networks were studied. Neo-fuzzy neurons were cho-
sen as nodes of the hybrid network, which allows to reduce computational costs. The
optimal network parameters were found. The synthesis of the optimal structure of
hybrid networks was performed. Experimental studies of LSTM, GMDH, and hy-
brid GMDH-neo-fuzzy networks with optimal parameters for short- and middle-
term forecasting have been conducted. The accuracy of the obtained experimental
predictions is compared. The forecasting intervals for which the application of the
researched artificial intelligence methods is the most expedient have been deter-
mined.
Keywords: optimization, GMDH, hybrid GMDH-neo-fuzzy network, LSTM, short-
and middle-term forecasting.
INTRODUCTION
Problems of forecasting share prices and market indexes at stock exchanges pay
great attention of investors and various money funds. For its solution were devel-
oped and for a long time applied powerful statistical methods, first of all ARIMA
[1; 2]. Last years different intelligent methods and technologies were also sug-
gested and widely used for forecasting in financial sphere, in particular among
them neural networks and fuzzy logic systems.
The efficient tool of modelling and forecasting of non-stationary time series
is Group method of data Handling (GMDH) suggested and developed by acad.
Alexey Ivakhnenko [3; 4]. This method is based on self-organization and enables
to construct optimal structure of forecasting model automatically in the process of
algorithm run. Methods GMDH and fuzzy GMDH were successfully applied for
forecasting at stock exchanges for long time.
As alternative approach for forecasting in finance is application of various
types of neural network: MLP [5], fuzzy neural networks [6; 7], neo-fuzzy net-
works [8] and Deep learning (DL) networks [9].
New trend in sphere DL networks is a new class of neural networks – hybrid
DL networks based on GMDH method [10]. The application of self-organization
Investigation of computational intelligence methods in forecasting at financial markets
Системні дослідження та інформаційні технології, 2023, № 3 55
in these networks enables to train not only neuron weights but to construct opti-
mal structure of a network. Due to a method of training in these networks weights
are adjusted not simultaneously but layer after layer. That prevents the phenome-
non of vanishing or explosion of gradient. It’s very important for networks with
many layers.
The first works in this field used as nodes of the hybrid network Wang-
Mendel neurons with two inputs [10]. But drawback of such neurons is the neces-
sity to train not only neural weights but the parameters of fuzzy sets in antece-
dents of rules as well. That needs a lot of calculation expenses and large training
time as well. Therefore, later DL neo-fuzzy networks were developed in which as
nodes were used neo-fuzzy neurons by Yamakawa [8; 11; 12]. The main property
of such neurons is that it’s necessary to train only neuron weights but not fuzzy
sets. That demands less computation in comparison to Wang-Mendel neurons and
significantly cuts training time as a whole. The investigation of both classes of
hybrid DL networks was performed and their efficiency at forecasting in financial
sphere was compared in [13].
At the same time for long term forecasting LSTM networks were developed
[14–16] and successfully applied for forecasting in economy and financial sphere.
LSTM networks have long memory where the information about preceding values
of forecasted time series is stored and they are enabled to forecast at middle term
and long term forecasting intervals. Therefore, it presents great interest to com-
pare the efficiency of hybrid DL networks, GMDH and LSTM at the problems of
short-term and middle-term forecasting at financial sphere.
The goal of this paper is to investigate the accuracy of intelligent methods –
hybrid DL networks, GMDH and LSTM at the problem of forecasting market in-
dices at the stock exchange at the different forecasting intervals (short-term and
middle-term), compare their efficiency and to determine the classes of forecasting
problems for which the application of corresponding computational intelligence
methods is the most perspective.
THE DESCRIPTION OF THE EVOLVING HYBRID GMDH-NEO-FUZZY
NETWORK
The evolving hybrid DL-network architecture is presented in Fig. 1. To the sys-
tem’s input layer a 1n -dimensional vector of input signals is fed. After that
x1
x2
xn
Fig. 1. Evolving GMDH-network
Yu. Zaychenko, He. Zaichenko, O. Kuzmenko
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 56
this signal is transferred to the first hidden layer. This layer contains 2
1 nn c
nodes, and each of these neurons has only two inputs.
At the outputs [1]N of the first hidden layer the output signals are formed.
Then these signals are fed to the selection block of the first hidden layer.
It selects among the output signals [1]
1 *ˆly n (where 1*n F is so-called
freedom of choice) most precise signals by some chosen criterion (mostly by the
mean squared error
2
1
ly
). Among these 1 *n best outputs of the first hidden
layer [1]
2 ˆ *ly n pairwise combinations [1] [1]*, ˆ *ˆl py y are formed. These signals are
fed to the second hidden layer, that is formed by neurons [2]N . After training
these neurons output signals of this layer [2]ˆly are transferred to the selection
block [2]SB which choses F best neurons by accuracy (e.g. by the value of 2
2
ly
)
if the best signal of the second layer is better than the best signal of the first hid-
den layer [1]
1ˆ *y . Other hidden layers work similarly. The system evolution proc-
ess continues until the best signal of the selection block [ 1]sSB appears to be
worse than the best signal of the previous s-h layer. Then it’s necessary to return
to the previous layer and choose its best node neuron [ ]sN with output signal
[ ]ˆ sy . And moving from this neuron (node) along its connections backwards and
sequentially passing all previous layers the final structure of the GMDH-neo-
fuzzy network is constructed.
It should be noted that in such a way not only the optimal structure of the
network may be constructed but also well-trained network due to the GMDH al-
gorithm. Besides, since the training is performed sequentially layer by layer the
problems of high dimensionality as well as vanishing or exploding gradient are
avoided.
NEO-FUZZY NEURON AS A NODE OF HYBRID GMDH-SYSTEM
Let’s consider the architecture of the node that is presented in Fig. 2 and is sug-
gested as a neuron of the proposed GMDH-system. As a node of this structure a
neo-fuzzy neuron (NFN) developed by Takeshi Yamakawa and co-authors in [9]
is used. The neo-fuzzy neuron is a nonlinear multi-input single-output system
shown in Fig. 2. The main difference of this node from the general neo-fuzzy neu-
ron structure is that each node uses only two inputs.
It realizes the following mapping:
2
1
ˆ ( )i i
i
y f x
,
where ix is the input ) , ,2 ,1 ( nii , ŷ is a system output. Structural blocks of
neo- fuzzy neuron are nonlinear synapses iNS which perform transformation of
input signal in the form
Investigation of computational intelligence methods in forecasting at financial markets
Системні дослідження та інформаційні технології, 2023, № 3 57
)()(
1
ijiji
h
j
ii xwxf
and realize fuzzy inference: if ix is jix then the output is jiw ,where jix is a fuzzy
set which membership function is ji , jiw is a synaptic weight in consequent [11].
THE NEO-FUZZY NEURON LEARNING ALGORITHM
The learning criterion (goal function) is the standard local quadratic error function:
.))(()(
2
1
)(
2
1
))(ˆ)((
2
1
)(
2
1
2
1
22
kxwkykekykykE ijiji
h
ji
It is minimized via the conventional stochastic gradient descent algorithm.
In case we have a priori defined data set the training process can be performed
in a batch mode at one epoch using conventional least squares method [12]
)()()()()()()()( ]1[
1
]1[]1[
1
]1[]1[
1
]1[ kykNPkykkkNw
N
k
N
k
T
N
k
,
where (•)+ means pseudo inverse of Moore–Penrose (here ( )y k denotes external
reference signal (real value).
h1
h2
w11
w21
wh1
w12
w22
wh2
x1
x2
f1(x1)
ŷ
f2(x1)
Fig. 2. Architecture of neo-fuzzy neuron with two inputs
Yu. Zaychenko, He. Zaichenko, O. Kuzmenko
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 58
If training observations are fed sequentially in on-line mode, the recurrent
form of the LSM can be used in the form:
.
))(()1()))(((1
)1()))(())((()1(
)1()(
,
))(()1()))(((1
))(()))(())1(()(()1(
)1()(
T
T
T
T
kxkΡkx
kΡkxkxkΡ
kΡkΡ
kxkΡkx
kxkxkwkykΡ
kwkw
ijijij
ijijijij
ijij
ijijij
ijijij
l
ij
ij
l
ij
l
DATASET
As the data set for forecasting were taken close values of market index NASDAQ
Composite in the period since 01.01.22 till 01.01.23. The whole sample consisted
of 251 instances included Open values, minimal, maximal and Close values and
volume in each day. The sample was divided into training and test subsamples.
The dynamics of NASDAQ Close values is shown in the Fig. 3.
The correlogram of NASDAQ index is presented in the Fig. 4.
Fig. 3. Dynamics of the index Close
Fig. 4. Correlogram
Investigation of computational intelligence methods in forecasting at financial markets
Системні дослідження та інформаційні технології, 2023, № 3 59
Analyzing the presented curve, one may conclude that there is strong corre-
lation between preceding and conceding values and even for lag 50 days the cor-
relation is more than 0.5.
EXPERIMENTAL INVESTIGATIONS
In the investigations was explored the forecasting accuracy of hybrid DL neo-
fuzzy networks at various forecasting intervals: short-term forecasting with inter-
vals 1, 3, 5 and 7 days and middle-term forecasting with intervals 20 and 30 days.
At the first step the variable experimental parameters of hybrid network were
chosen which are presented in the Table 1.
T a b l e 1 . Experimental parameters
Parameter Value
Membership functions Gaussian
Number of inputs 3; 4; 5
Number of linguistic variables 3; 4; 5
Ratio (percentage of the training sample) 0.6 (60%); 0.7 (70%); 0.8 (80%)
Criterion MSE; MAPE
Forecast interval 1; 3; 5; 7; 20; 30
The optimization of these parameters was performed in result the following
optimal values were determined inputs: 3; linguistic variables: 3; ratio: 0.7.
After that the structure optimization of hybrid DL neo-fuzzy network was
performed using GMDH method. The process of structure generation is presented
in the Table 2.
T a b l e 2 . Structure generation (inputs: 3; variables: 3; ratio: 0.7)
Nodes SB1 SB2 SB3
(0, 1) 2.6152319
(0, 2) 5.6112545
(1, 2) 3.8828252
((0, 1), (0, 2)) 0.03519317
((0, 1), (1, 2)) 0.0357832
((0, 2), (1, 2)) 0.05844182
(((0, 1), (0, 2)), ((0, 1), (1, 2))) 0.09281185
(((0, 1), (0, 2)), ((0, 2), (1, 2))) 0.11276198
(((0, 1), (1, 2)), ((0, 2), (1, 2))) 0.08893768
In result the optimal structure of three layers: at the first layer 3 inputs, sec-
ond layer – two neurons, third layer – one output neuron.
Further the training of the best hybrid network was carried out using method
SGD (stochastic gradient descent) with variable step. Flow chart of forecasting
results for interval 20 in presented in the Fig. 5. The values of MSE and MAPE
for this experiment are shown in the Table 3.
Yu. Zaychenko, He. Zaichenko, O. Kuzmenko
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 60
T a b l e 3 . Forecasting accuracy of hybrid neo-fuzzy network at forecasting
interval 20 days
Criterion MSE MAPE
min 30.68518 0.049986
average 158515.7 3.024738
maximal 811272.4 8.818966
In the Fig. 6. flow chart of MAPE values for the best model of hybrid net-
work is shown.
Further the similar experiments of hybrid network were performed with
forecasting interval 30 days. After optimization the parameters and structure of
hybrid network it was trained using training subsample. The forecasting accuracy
y at the test sample is presented at the Table 4.
Fig. 6. MAPE for the best forecast (inputs: 3; variables: 3; ratio: 0.7)
Fig. 5. The best forecast (inputs: 3; variables: 3; ratio: 0.7)
Investigation of computational intelligence methods in forecasting at financial markets
Системні дослідження та інформаційні технології, 2023, № 3 61
T a b l e 4 . Forecasting accuracy of hybrid neo-fuzzy network at interval 30 days
Criterion MSE MAPE
min 177.865 0.120699
average 164611 3.07087
maximal 840641.8 8.977178
For estimating forecasting accuracy of hybrid DL network, it was compared
with alternative methods: GMDH and LSTM. For GMDH algorithm the follow-
ing parameters values were set after preliminary explorations: linear partial de-
scriptions, number of inputs 5, ratio training/test 0.6. Flow chart of the best fore-
cast is shown in the Fig. 7 and Fig. 8.
After that the experiments were performed with LSTM network. LSTM was
trained and tested at the different forecasting intervals 1, 3, 5, 7, 20 and 30 days.
The goal of experiments was to find the optimal parameters. The following pa-
rameters varied: number of inputs 3-5, ratio training/test 0.6, 0.7, 0.8. After that
the LSTM with optimal parameters was applied for forecasting.
Fig. 7. The best forecast (inputs: 3; variables: 3; ratio: 0.8) for interval 30 days
Fig. 8. The best forecast by GMDH (function: linear; inputs: 5; ratio: 0.6) 20 days
Yu. Zaychenko, He. Zaichenko, O. Kuzmenko
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 62
In the Table 5 forecasting accuracy of LSTM network at interval 3 days and
in the Fig. 9 forecasting results are presented. The optimal parameters values were
found number of inputs 5, ratio training/test 0.6.
T a b l e 5 . Forecasting accuracy of LSTM network at forecasting interval 3 days
Criterion MSE MAPE
min 113.4100292 0.098063438
average 117981.36 2.652192244
maximal 517650.7403 6.953914724
The values of MSE and MAPE for forecasting with an interval of 20 days
are shown in Table 6. The forecasting results are presented in Fig. 10.
T a b l e 6 . Forecasting accuracy of LSTM network at forecasting interval 20 days
Criterion MSE MAPE
min 49.56215352 0.06300144
average 327754.696 4.11679646
maximal 1545745.838 12.17316133
Fig. 9. The best forecast by LSTM (inputs: 5; ratio: 0.6) 3 days
Fig. 10. The best forecast by LSTM (inputs: 5; ratio: 0.6) 20 days
Investigation of computational intelligence methods in forecasting at financial markets
Системні дослідження та інформаційні технології, 2023, № 3 63
The comparative experiments were performed in which the accuracy of fore-
casting by hybrid DL network, GMDH and LSTM at the different forecasting in-
tervals was estimated and compared. The corresponding results are presented in
the Tables 7, 8 and Fig. 11, 12.
T a b l e 7 . Average MSE values of the best models for different intervals
Іnterval GMDH-neo-fuzzy GMDH LSTM
interval 1 97865.41363 44462.69 55461.3459
interval 3 104012.245 122615 117981.36
interval 5 155308.7139 151131.5 220850.108
interval 7 156023.0308 191982.4 241535.576
interval 20 158515.6721 243991.7 327754.7
interval 30 164610.9742 245615.6 327216.9
T a b l e 8 . Average MAPE values of the best models for different intervals
Іnterval GMDH-neo-fuzzy GMDH LSTM
interval 1 2.483877618 1.557535 1.76242389
interval 3 2.544556353 2.623422 2.65219224
interval 5 2.889892779 3.035898 3.56067021
interval 7 2.867433998 3.428108 3.73361624
interval 20 3.02473808 3.710976 4.116796
interval 30 3.070870375 3.870127 4.25219
Fig. 11. Average MSE values of the best models for different intervals
Fig. 12. Average MAPE values of the best models for different intervals
Yu. Zaychenko, He. Zaichenko, O. Kuzmenko
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 64
Analyzing the presented results in the Fig. 11 one may conclude that GMDH
method appears to be the best at short term forecasting 1, 3 days which complies
the theory.
Hybrid deep learning neo-fuzzy networks are the best at middle-term fore-
casting 7, 20, 30 days. LSTM networks appeared to be the worst by accuracy as
compared with intelligent methods – hybrid DL networks and GMDH.
CONCLUSION
In this paper the investigations of artificial intelligence methods: hybrid Deep
learning networks and GMDH were carried out in the problem of forecasting
NASDAQ close prices.
During the experiments the optimal structure and optimal parameters: num-
ber of inputs, number of linguistic values, ratio training/test samples of hybrid
neo-fuzzy networks were determined.
After optimization of hybrid neo-fuzzy networks and parameters of GMDH
method the experiments on forecasting NASDAQ Close were performed at different
intervals: 1, 3, 5, 7 (short-term forecast) and 20, 30 days (middle-term forecast).
The accuracy of forecasting by Hybrid DL networks and GMDH was com-
pared with alternative method – LSTM networks.
The analysis of obtained results have shown that GMDH method is the best
at short term forecasting 1, 3 days while hybrid deep learning neo-fuzzy networks
are the best at middle-term forecasting 7, 20, 30 days. LSTM networks appeared
to be the worst by accuracy as compared with intelligent methods – hybrid DL
networks and GMDH.
REFERENCES
1. Peter J. Brockwell and Richard A. Davis, Introduction to time series and forecasting;
2nd ed. Springer, 2002, 429 p.
2. Robert H. Shumway and David S. Stoffer, Time Series Analysis and its Applications
with R Examples; 4-th edition. Springer, 2017, 562 p.
3. A.G. Ivakhnenko, G.A. Ivakhnenko, and J.A. Mueller, “Self-organization of the neu-
ral networks with active neurons,” Pattern Recognition and Image Analysis, vol. 4,
no. 2, pp. 177–188, 1994.
4. A.G. Ivakhnenko, D. Wuensch, and G.A. Ivakhnenko, “Inductive sorting-out GMDH
algorithms with polynomial complexity for active neurons of neural networks,”
Neural Networks, 2, pp. 1169–1173, 1999.
5. S.S. Haykin, Neural networks: a comprehensive foundation; 2nd ed. Upper Saddle
River, N.J: Prentice Hall, 1999.
6. S. Ossovsky, Neural networks for information processing. M.: Finance and Statis-
tics, 2002, 344 p.
7. F. Wang, “Neural Networks Genetic Algorithms and Fuzzy Logic for Forecasting,”
Proc. Intern. Conf. Advanced Trading Technologies. New York, 1992, pp. 504–532.
8. T. Yamakawa, E. Uchino, T. Miki, and H. Kusanagi, “A neo-fuzzy neuron and its
applications to system identification and prediction of the system behavior,” Proc.
2nd Intеrn. Conf. Fuzzy Logic and Neural Networks «LIZUKA-92», Lizuka, 1992,
pp. 477–483.
9. I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. MIT PRESS, 2016.
Available: http://www.deeplearningbook.org
Investigation of computational intelligence methods in forecasting at financial markets
Системні дослідження та інформаційні технології, 2023, № 3 65
10. Yuriy Zaychenko, Yevgeniy Bodyanskiy, Oleksii Tyshchenko, Olena Boiko, and
Galib Hamidov, “Hybrid GMDH-neuro-fuzzy system and its training scheme,” Int.
Journal Information theories and Applications, vol. 24, no. 2, pp. 156–172, 2018.
11. Yu. Zaychenko and Galib Hamidov, “The Hybrid Deep Learning GMDH-neo-fuzzy
Neural Network and Its Applications,” Proceedings of 13-th IEEE Intern Conference
Application of Information and Communication Technologies-AICT2019, 23–25 Oc-
tober 2019, Baku, pp. 72–77.
12. Evgeniy Bodyanskiy, Yuriy Zaychenko, Olena Boiko, Galib Hamidov, and Anna
Zelikman, “Structure Optimization and Investigations of Hybrid GMDH-Neo-fuzzy
Neural Networks in Forecasting Problems,” System Analysis & Intelligent Comput-
ing; eds. Michael Zgurovsky, Natalia Pankratova (Book Studies in Computational
Intelligence, SCI), vol. 1022. Springer, 2022, pp. 209–228.
13. Yuriy Zaychenkoa, Helen Zaichenkoa, and Galib Hamidov, “Hybrid GMDH Deep
Learning Networks – Analysis, Optimization and Applications in Forecasting at Fi-
nancial Sphere,” System Research and Information Technologies, no. 1, pp. 73–86,
2022. doi: 10.20535/SRIT.2308-8893.2022.1.06.
14. S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation,
vol. 9, pp. 1735–1780, 1997. doi: 10.1162/neco.1997.9.8.1735.
15. B. Hammer, “On the approximation capability of recurrent neural networks,” Neuro-
computing, vol. 31, pp. 107–123, 1998. doi: 10.1016/S0925-2312(99)00174-5.
16. C. Olah, Understanding LSTM networks, 2020. Available: https://colah.github.io/
posts/2015-08-Understanding-LSTMs/
17. A. Graves, “Generating sequences with recurrent neural networks,” CoRR, vol.
abs/1308.0850, 2013. doi: 10.48550/arXiv.1308.0850.
Received 08.05.2023
INFORMATION ON THE ARTICLE
Yuriy P. Zaychenko, ORCID: 0000-0001-9662-3269, Educational and Research
Institute for Applied System Analysis of the National Technical University of
Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: zaychenkoy-
uri@ukr.net
Helen Yu. Zaichenko, ORCID: 0000-0002-4630-5155, Educational and Research Insti-
tute for Applied System Analysis of the National Technical University of Ukraine “Igor
Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: syncmaster@bigmir.net
Oleksii V. Kuzmenko, ORCID: 0000-0003-1581-6224, Educational and Research Insti-
tute for Applied System Analysis of the National Technical University of Ukraine “Igor
Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: oleksii.kuzmenko@ukr.net
ДОСЛІДЖЕННЯ МЕТОДІВ ОБЧИСЛЮВАЛЬНОГО ІНТЕЛЕКТУ У
ПРОГНОЗУВАННІ НА ФІНАНСОВИХ РИНКАХ / Ю.П. Зайченко, О.Ю. Зай-
ченко, О.В. Кузьменко
Анотація. Розглянуто інтелектуальні методи для короткострокового та серед-
ньострокового прогнозування у фінансовій сфері. Досліджувалися DL мережі
LSTM, МГУА та гібридні МГУА неофаззі мережі. Як вузли гібридної мережі
обрано неофаззі нейрони, що дозволяє зменшити обчислювальні витрати.
Знайдено оптимальні параметри мереж. Виконано синтез оптимальної струк-
тури гібридних мереж. Проведено експериментальні дослідження мереж
LSTM, МГУА та МГУА неофаззі з оптимальними параметрами для коротко-
строкового та середньострокового прогнозування. Порівняно точність отрима-
них експериментальних прогнозів. Визначено інтервали прогнозування, для
яких застосування досліджених методів штучного інтелекту є найбільш
доцільним.
Ключові слова: оптимізація, МГУА, гібридна мережа МГУА-неофаззі, LSTM,
короткострокове та середньострокове прогнозування.
|
| id | journaliasakpiua-article-290368 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:28:21Z |
| publishDate | 2023 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/d2/c18f43e918749f836e2afa30528eded2.pdf |
| spelling | journaliasakpiua-article-2903682023-11-07T22:19:24Z Investigation of computational intelligence methods in forecasting at financial markets Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках Zaychenko, Yuriy Zaichenko, Helen Kuzmenko, Oleksii оптимізація МГУА гібридна мережа МГУА-неофаззі LSTM короткострокове та середньострокове прогнозування optimization GMDH hybrid GMDH-neo-fuzzy network LSTM short- and middle-term forecasting The work considers intelligent methods for solving the problem of short- and middle-term forecasting in the financial sphere. LSTM DL networks, GMDH, and hybrid GMDH-neo-fuzzy networks were studied. Neo-fuzzy neurons were chosen as nodes of the hybrid network, which allows to reduce computational costs. The optimal network parameters were found. The synthesis of the optimal structure of hybrid networks was performed. Experimental studies of LSTM, GMDH, and hybrid GMDH-neo-fuzzy networks with optimal parameters for short- and middle-term forecasting have been conducted. The accuracy of the obtained experimental predictions is compared. The forecasting intervals for which the application of the researched artificial intelligence methods is the most expedient have been determined. Розглянуто інтелектуальні методи для короткострокового та середньострокового прогнозування у фінансовій сфері. Досліджувалися DL мережі LSTM, МГУА та гібридні МГУА неофаззі мережі. Як вузли гібридної мережі обрано неофаззі нейрони, що дозволяє зменшити обчислювальні витрати. Знайдено оптимальні параметри мереж. Виконано синтез оптимальної структури гібридних мереж. Проведено експериментальні дослідження мереж LSTM, МГУА та МГУА неофаззі з оптимальними параметрами для короткострокового та середньострокового прогнозування. Порівняно точність отриманих експериментальних прогнозів. Визначено інтервали прогнозування, для яких застосування досліджених методів штучного інтелекту є найбільш доцільним. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2023-09-29 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/290368 10.20535/SRIT.2308-8893.2023.3.04 System research and information technologies; No. 3 (2023); 54-65 Системные исследования и информационные технологии; № 3 (2023); 54-65 Системні дослідження та інформаційні технології; № 3 (2023); 54-65 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/290368/283955 |
| spellingShingle | оптимізація МГУА гібридна мережа МГУА-неофаззі LSTM короткострокове та середньострокове прогнозування Zaychenko, Yuriy Zaichenko, Helen Kuzmenko, Oleksii Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках |
| title | Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках |
| title_alt | Investigation of computational intelligence methods in forecasting at financial markets |
| title_full | Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках |
| title_fullStr | Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках |
| title_full_unstemmed | Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках |
| title_short | Дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках |
| title_sort | дослідження методів обчислювального інтелекту у прогнозуванні на фінансових ринках |
| topic | оптимізація МГУА гібридна мережа МГУА-неофаззі LSTM короткострокове та середньострокове прогнозування |
| topic_facet | оптимізація МГУА гібридна мережа МГУА-неофаззі LSTM короткострокове та середньострокове прогнозування optimization GMDH hybrid GMDH-neo-fuzzy network LSTM short- and middle-term forecasting |
| url | https://journal.iasa.kpi.ua/article/view/290368 |
| work_keys_str_mv | AT zaychenkoyuriy investigationofcomputationalintelligencemethodsinforecastingatfinancialmarkets AT zaichenkohelen investigationofcomputationalintelligencemethodsinforecastingatfinancialmarkets AT kuzmenkooleksii investigationofcomputationalintelligencemethodsinforecastingatfinancialmarkets AT zaychenkoyuriy doslídžennâmetodívobčislûvalʹnogoíntelektuuprognozuvannínafínansovihrinkah AT zaichenkohelen doslídžennâmetodívobčislûvalʹnogoíntelektuuprognozuvannínafínansovihrinkah AT kuzmenkooleksii doslídžennâmetodívobčislûvalʹnogoíntelektuuprognozuvannínafínansovihrinkah |