CC BY
19TECHNICAL SCIENCES
APPLICATION OF ARTIFICIAL NEURAL NETWORK FOR FORECASTING INTERNATIONAL
FINANCIAL MARKETS
Cozac E.
Masters Degree in Computer Science Software Engineer GAN C/O Memery Crystal Llp, London, United Kingdom
ABSTRACT
The article reveals the principles of using an artificial neural network to forecast international financial markets. It is emphasized that the use of artificial neural networks for forecasting international financial markets is based on the high ability to generalize, which helps to improve the accuracy of the forecast. The peculiarities of international financial markets and the sequence of formation of neural networks for analysis and forecasting are described. It is emphasized that the possibility of learning is the main advantage of neural networks over traditional algorithms of financial forecasting, technical training is to find the coefficients of connections between neurons. It is noted that the ability of the neural network to predict directly follows from its ability to generalize and highlight the hidden relationships between input and output data. The process of learning an artificial neural network with the separation of the main signals is shown schematically. A description of the stages of formation, which are mandatory for the formation of an artificial neural network for forecasting international financial markets. The mathematical model of the financial time series, which is the basis of the forecasting process, is presented and it is emphasized that the task of forecasting is possible to present as a search function. The process of determining weights is described, it is noted that the artificial neural network selects weights according to training data and checks training results on validation data, if the network prediction error on test data increases, the neural network is retrained and training data is stored. It is emphasized that the functions of cross-entropy, exponential or exponential function and function of information germination are used as a function of losses, and the choice of function depends on the indicators of the impact on the financial series. The advantages of artificial neural networks in the framework of international financial market forecasting are described and it is emphasized that the forecast of financial time series is the main activity of the entire investment industry - all exchanges and over-the-counter securities trading systems, especially internationally. The resulting fact is that given that the financial time series is a continuous function, the use of neural networks is quite justified and correct for accurate forecasting.
Keywords: forecasting, international financial market, artificial neural network, economy, time series, error, signal.
Introduction and task. Modern economic space all with greater acuteness is of interest in qualitative forecasting of international financial markets, taking into account the rapid growth of digital technologies and innovative development of data analysis tools. However, taking into account the experience of previous years, a technical analysis that used most market participants is not effective, the use of standard mathematical analysis does not give the desired result, since the economy is often irrational because it moves irrational motivations of people. In recent years, financial analysts began to trigger artificial neural networks that are mathematical models, as well as their software or hardware implementations built on the principle of organization and functioning of biological neural networks.
Artificial neural networks as innovative technologies in the framework of financial, analytical and mathematical analysis are considered as a mechanism for the reconstruction of a decision maker, and due to its relevant characteristics can be used to predict the profitability in the international financial market.
The thorough problem of studies of artificial neural networks, their formation and increase of the forecast degree is to take into account various processing input data methods. This is because in recent years of data arrays available to financial analysts, significantly
increased, and interconnections between them, in particular, causal complicated. Therefore, the process of establishing the effects of influential factors and the training of the network based on a heterogeneous sample is important.
Literature review. In recent years, a number of works are published, which investigates the main aspects of the use of an artificial neural network in various spheres of life, its development, formation and innovative discovery.
N. Ya. Savka [1] developed an architecture of an artificial neural network with radially baseline functions for modeling and forecasting financial security indexes. The author states that one of the main factors influencing financial security indicators is tax debt. The results of experiments are confirmed by the effectiveness of the architecture of the neural network of the radial type.
O. Pastukh, I. Kramar and O. Chernih [2] evaluated the effectiveness of industrial enterprises in the process of their internationalization with means of artificial intelligence. Scientists have investigated the benefits of using neural modeling method to solve economic tasks and substantiated the expediency of its use in conducting this study.
O. L. Burleyev, O. O., Vasilenko and R. M, Ivanenko [3] considered the peculiarities of the artificial neural networks creation, their training, application in the economic sphere and comparing their effectiveness with statistical methods. It is established that neural networks are used to solve the three main types of tasks: forecasting, classification and modeling. Platforms and libraries helping when creating a neural network and have ready-made samples of use and detailed documentation. It is confirmed that the main advantages of neural networks are the ability to study, the ability to work with incomplete data, the ability to automate an analysis, high accuracy of results.
The use of artificial neural networks in the system of risk management of the enterprise is discussed in [4]. As part of the study, the risk is considered as an unwanted event that translates the object of management into an undesirable state. It is proved that the use of artificial neural network models will allow you to move from traditional jet management in response to the risk events to proactive, which involves managerial impacts on the causes of risks and indirectly on negative events and their consequences.
N. N. Paid-Nosik and G. V. Mazyutinets [5] described the application principles of artificial neural networks to analyze the level of company's financial security. M.O. Mozolevskaya and O.V. Stavitsky [6] investigated the specifics of the neural networks use for prediction in the financial sector.
From foreign authors it is worth noting such works as: Siriopoulos Costas & Markellos Raphael & Sir-lantzis Konstantinos[7], Armutlulu ismail & Serhad-lioglu Gurkan [8], Moon Kaleem & Siddiqui Shahan & Shoaib Bilal & Abbas Irfan & Khan Abdul & Irshad Muhammad & Farooq Muhammad[9], Gorekore Kelvin [10],Okasha Mahmoud & Yassen Assem [11], Del Rosso Maria Pia & Ullo Silvia & Sebastianelli Ales-sandro & Spiller Dario & Puglisi Erika & Biondi Fil-ippo & Orlando Danilo [12], Egrioglu Erol & Yolcu Ufuk & Bas Eren & Dalar Ali[13], Marzi Hosein & Turnbull Mark [14],Maeda Iwao & Matsushima Hi-royasu & Sakaji Hiroki & Izumi Kiyoshi & deGraw David & Kato Atsuo & Kitano Michiharu [15], Khan Say-yid Iskandar [16]and others.
However, taking into account the described scientific nurses, on the topic, the analysis of the principles of application of an artificial neural network to predict international financial markets remains open and requires detailed processing.
Objectives. The main research aim is to analyze the principles of an artificial neural network application to predict international financial markets.
Main research results. The artificial neural networks use to predict international financial markets is based on high levels of generalization, which contributes to increasing the forecast accuracy.
Financial markets are known as complex systems that do not provide for deterministic links between the current and future markets. Therefore, the forecast for
the future period should be performed taking into account uncertainty. In the field of financial market for future, it can be used to facilitate the decision-making processes associated with investment behavior. The market followersstrategy and the opposites strategy is one of the simplest approaches to making decisions based on prediction. However, in the case where the forecasting of the financial market includes uncertainty, decision-making based on uncertain predictions may result in losses.
Unlike other algorithms for forecasting, artificial neural networks are not programmed, and self-accomplished in the work process. Therefore, it can be expected that they will give good models for financial time rows, which are international financial markets and, accordingly, provide the best forecasts for future periods.
Since all artificial neural networks are based on the concept of neurons, connections and transmission functions, there is similarity between different neural networks structures. Most differences depend on different learning methods. For the learning process, it is necessary to have a model of the external environment in which the neural network operates - the necessary information for solving the problem. One must also determine how to change the network weight parameters. The network training refers to the procedure that uses learning rules for weight adjustment.
The neural network can study with or without a teacher. After re-representation, the weight of the neural network is stabilized, and the neural network gives the correct answers to all (or almost all) of the program from the database. In software implementations, one can see that the error value (the sum of the error for all outputs) gradually decreases in the learning process.
When the error value reaches zero or an acceptable low level, learning is stopped and the received neural network is considered to be trained and ready to use new data. One of the most significant areas of application of neural networks in the financial sector is forecasting in the stock market.
The ability to teach is the main advantage of neural networks before traditional algorithms. Technically training is finding the coefficients of connections between neurons. In the learning process, the neural network is capable of detecting complex relationships between the input data and output, as well as to perform generalization. The ability of the neural network to prediction directly derived from its ability to synthesize and allocate hidden dependencies between the input and output data. In the case of successful training, the neural network is capable of predict the future value of a certain sequence based on several previous values and / or some factors existing at the time of training. It should be noted that prediction is possible only when the previous changes are definitely determined to some extent. The process of teaching an artificial neural network is shown in Figure 1.
Figurel - The teaching process of an artificial neural network
Formation of an artificial neural network for the forecasting of international financial markets is due to the execution of formation successive stages (Figure 2): The first stage is characterized by preparing an array of data that is subject to analysis and forecast. Data and exit data are allocated. The main indicators of the financial market of any country are its GDP, the oil cost, the national currency cost, euro and dollars, gold stability and valuable metals in the market, etc.
The second stage is a calibration of an array of data that is formed and directed to analysis. When calibrat-
and the result of the forecast compares to the actual value, with an unsatisfactory comparison, the network changes the configuration of its own parameters so that the following results are reduced to minimizing the error on the output.
The next, third stage is to establish a deadline for the forecast, that is, the definition of time space indicating the stop time.
The latest stage is to obtain the result of an artificial neural network that is, obtaining a predictable value at the initial sample of financial indicators that are
ing a number of data, an artificial neural network ana- formed in the database as a starting set of values. lyzes the obtained data gives the preliminary prediction
Figure 2- Blockdiagram of the stages of forming an artificial neural network for the implementation of
international financial markets forecasting
The mathematical model of the financial time series is the basis of the forecasting process. Prediction task is the ability to introduce as a search function:
xT+k ~ fr(xT, ■■■,xi,k) = %T+kiT
wherefc 6 {1,„.,K} - postponing the forecast of a financial row;
K - Horizon of the financial row forecast.
Delaying the financial rank forecast determines how the time interval of the forecast will be broad. To carry out the forecasting of the financial rank, a autoregression model and a model of sliding average, as well as their generalization are used.
Auto regression Model1:
X, = c +
^ ViXt-i + £t
where^,..., <pp - Financial Row Parameters; c - constant; et - error.
Sliding average model:
q
Xt= v + + ^ @i£t-i i=1
Where^, ...,9q - the financial row model parameters;
H - expected value; et - error.
The input data formation is to normalize the initial row by the range of values:
y — y X = -
wherex - unit of financial row;
, xn
^max* ^mm
- maximum and minimum value of a
financial row accordingly.
After such a transformation, each value of the financial rank consisting of n consecutive values is normalized so that all the values of the financial row lie in the range of 0 to 1. In this case, the real values are lost, and the records are in hypercub[0,1]„. Thus, at any level of financial rank indicators guarantees the invariance of the input recording.
Standardization of the financial rank is to scale the sets of values in such a way that the average value -equal to zero, and the value of the standard deviation was equal to the unit. Standardized values are calculated by formula:
x — X
whereX - financial row average value;
ax - the magnitude of the standard deviation.
Artificial neural network selects weight coefficients in accordance with the learning data and checks the results of training on validation data. If the network forecast error on test data increases, then the release of the neural network and the learning data are remembered.
To calculate an error indicator, a loss function is used. Adjustment of weight coefficients is carried out using the following formula:
Wk+i =wk- aVwE(w)
wherea - Network Training speed (numerical coefficient);
E(w)- Loss function;
Vw- Operator Hamilton;
wk - Weight indicator.
To calculate the new weight coefficients, a differential Hamilton operator is used, indicating a gradient of loss function. As a function of losses, the function of the mean square error is most often used:
n
E = 1~Jj(yl-yl)2
L = 1
wherey - artificial neural network results;
y- training sample.
Also, the functions of cross-tension, exponential or indicator function and the function of the introduction of information are used. The choice of function depends on the indicators of impact on the financial row.
Artificial neural networks for predicting the international financial market have the following advantages:
- ease of use, since neural networks are studying in the examples of the resulting data. The user of the neural network selects representative data, and then launches a learning algorithm that automatically perceives the data structure;
- neural networks have advantages from an intuitive point of view, as they are based on a primitive biological model of the nervous system;
- prediction of financial time series is a necessary element of any investment activity. The very idea of investing in order to receive income in the future is based on the idea of prediction of the future.
The forecast of financial time series is the main activity of the entire investment industry - all exchanges and out-of-the-counter securities trading systems especially on international scale.
Neural networks allow:
- technical analysis of the initial data sample, depending on the timing;
- technical and fundamental analysis according to various parameters of a financial time series: a row, quotes, prices, refinancing rate, index of domestic gross product, financial indicators of the research object (country, city, company), index.
Conclusions. The paper reveals the application principles of an artificial neural network to predict international financial markets. Taking into account the fact that artificial neural networks are nonlinear in its entirety, neural network allow with any degree of accuracy to approximate an arbitrary continuous function, despite the absence or presence of any periodicity or cyclist. Taking into account that the financial time series is a continuous function, the use of neural networks is fully justified and correctly for accurate forecast.
i=i
1 http s://surl.li/bunjq
Prospects for further work are the development of an intellectual system for predicting economic activity of an enterprise based on artificial neural networks.
References
1. Savka N.Ya. (2019). Artificial neural networks with radially baseline functions for modeling financial security indicators. Inductive modeling of complex systems: Sb. Science, 11, P. 95-102.
2. Shepherd O. (2019). Using neural networks to ensure the effectiveness of industrial enterprises in the process of their internationalization. Galitsky Economic Bulletin. T.: TNTU, Tom 58, 3, P. 121-129.
3. Burleyev, O. L., Vasilenko, O., Ivanenko R. M. (2021). Efficiency of the use of artificial neural networks in the economy. Economics and society, 31.
4. Denisov, A., Richkova, L. (2021). Use of artificial neural networks in the risk management system. Electronic scientific professional publication "Adaptive Management: Theory and Practice" Series "Economics", 11 (22).
5. Spido-Nosik N.N., Mazuutinets G.V. (2020). Application of artificial neural networks to analyze the level of financial security companies. Scientific Bulletin of Uzhgorod University. Series Economics, 1 (55), 112-117.
6. Mozolevskaya M.O., Stavitsky O.V. (2017). Using neural networks to predict in the financial sector. Actual Problems of Economics and Management, 11, URL: http://ape.fmm.kpi.ua/article/view/102584.
7. Siriopoulos, Costas & Markellos, Raphael & Sirlantzis, Konstantinos. (2022). Applications of artificial neural networks in emerging financial markets. -URL: https://www.researchgate.net/publica-tion/266273741_Applications_of_artificial_neu-ral_networks_in_emerging_financial_markets
8. Armutlulu, ismail &Serhadlioglu, Gürkan. (2020). A purposed application of artificial neural networks in financial forecasting. Öneri Dergisi. 23-27. 10.14783/maruoneri.733012.
9. Moon, Kaleem & Siddiqui, Shahan & Shoaib, Bilal & Abbas, Irfan & Khan, Abdul & Irshad, Muhammad & Farooq, Muhammad. (2020). Study of Applications of Artificial Neural Networks. 4. 1-11.
10. Gorekore, Kelvin. (2022). Artificial neural networks in complex financial markets. - URL: https://www.researchgate.net/publication/265145629_
11. Okasha, Mahmoud & Yassen, Assem. (2013). The Application of Artificial Neural Networks In Forecasting Economic Time Series. International Journal of Statistics and Analysis (IJSA). 3. 257-277.
12. Del Rosso, Maria Pia & Ullo, Silvia & Sebas-tianelli, Alessandro & Spiller, Dario & Puglisi, Erika & Biondi, Filippo & Orlando, Danilo. (2021). Artificial neural network. 10.1049/PBTE098E_ch4.
13. Egrioglu, Erol & Yolcu, Ufuk & Bas, Eren & Dalar, Ali. (2019). Median-Pi artificial neural network for forecasting. Neural Computing and Applications. 31. 1-10. 10.1007/s00521-017-3002-z.
14. Marzi, Hosein & Turnbull, Mark. (2007). Use of Neural Networks in Forecasting Financial Market. Proceedings - 2007 IEEE International Conference on Granular Computing, GrC 2007. 516-516. 10.1109/GrC.2007.78.
15. Maeda, Iwao & Matsushima, Hiroyasu & Sa-kaji, Hiroki & Izumi, Kiyoshi & deGraw, David & Kato, Atsuo & Kitano, Michiharu. (2021). Predictive Uncertainty in Neural Network-Based Financial Market Forecasting. International Journal of Smart Computing and Artificial Intelligence. 5. 10.52731/ijscai.v5.i1.541.
16. Khan, Sayyid Iskandar. (2020). Application of neural network for trading financial markets. 10.13140/RG.2.2.10035.04643. - URL: https://www.researchgate.net/publication/346987383_ APPLICATION_OF_NEURAL_NETWORK_FOR_T RADING_FINANCIAL_MARKETS
HACCP SYSTEM IN THE RESTAURANT INDUSTRY
Blahopoluchna A.,
Pavlo Tychyna Uman State Pedagogical University, Lecturer-trainee,
Ukraine Liakhovska N.,
Uman National University of Horticulture., Senior Lecturer,
Ukraine Povorozniuk I.,
Pavlo Tychyna Uman State Pedagogical University, Associate Professor,
Ukraine
Barvinok N.
Pavlo Tychyna Uman State Pedagogical University, Senior Lecturer,
Ukraine
ABSTRACT
The article is devoted to the problem of implementation of HACCP systems in restaurants of Ukraine for compliance with current legislation and gaining competitive advantages in the hospitality industry. Keywords: HACCP system, restaurant business, food safety, critical control points.
