EFFICIENCY OF USING ARTIFICIAL NEURAL NETWORKS IN THE ECONOMY
Abstract
Much research has been devoted to neural networks, however there is little information available on effectiveness of different architectures in economics. The article investigates the peculiarities of artificial neural networks creation, their training, application in economics, and compares their effectiveness by statistical methods. A comparative study of the features of neural network kernels based on the support vector machine method with other methods for data classification shows that the support vector machine method is effective data classification tasks, especially for data with non-linear structure. Therefore, neural networks are really effective for the analysis of economic indicators and already far ahead of classical methods of analysis. Neural networks are used to solve three main types of tasks: forecasting, classification and modeling. Presented in the article are platforms and libraries, that help in creating a neural network and have ready-to-use samples and detailed documentation. The main advantages of neural networks are the ability to learn, the ability to work with incomplete data, the ability to automate the analysis, the high accuracy of the results. The main disadvantages of neural networks are the technical requirements, the need for a large amount of collected and processed data for training and the complexity of implementation in each individual case. The most common types of neural networks and training algorithms are presented, as well as economical tasks that are solved effectively by different types of neural networks. Performance of neural network perceptron and logistic regression were tested on the same data. These approaches have offered generally similar level of accuracy when solving the same test classification problem, with perceptron showing 3,5% more accurate result. In conclusion it can be said that artificial neural networks may offer more accurate result compared to classical methods of data analysis. Accuracy depends on type of task, amount and type of available data, and the complexity of the relationship between data.
References
Новікова В.В. Прогнозування внутрішнього валового продукту на основі апарату штучних нейронних мереж. Економіка та управління АПК. 2014. № 1. С. 5–11.
Курочкина И.П., Калинин И.И., Маматова Л.А., Шувалова Е.Б. Нейронные модели в диагностике финансового результата предприятий жилищно-коммунального хозяйства. Статистика и математические методы в экономике. 2019. № 3. С. 52–60.
Каргин Б.Б. Новые подходы к решению информационных задач в экономике и коммерции. Theoretical and Practical Aspects of Economics and Intellectual Property. 2017. № 16. С. 235–240.
Максимова О.М., Стёпина Е.Н. Использование нейронных сетей для решения задач строительной механики. URL: https://core.ac.uk/download/pdf/38644194.pdf/. (дата звернення: 14.07.2021).
Науменко В.А. Применение нейронных сетей для решения практических задач в экономике. Вектор экономики. 2019. № 10. С. 1–12.
Григорьева Д.Р. Применение пакета statistica для анализа и прогнозирования экономических показателей предприятия. URL: https://kpfu.ru/portal/docs/F2101409340/Grigoreva.pdf. (дата звернення: 10.08.2021).
Novikova V.V. (2014) Prohnozuvannia vnutrishnoho valovoho produktu na osnovi aparatu shtuchnykh neironnykh merezh [Prediction of gross domestic product based on the unit of piece neural fences]. Ekonomika ta upravlinnia APK, № 1, S. 5–11. (in Ukrainian)
Kurochkina I.P., Kalinin I.I., Mamatova L.A., Shuvalova E.B. (2019) Nejronnye modeli v diagnostike finansovogo rezul'tata predprijatij zhilishhno-kommunal'nogo hozjajstva [Neural models in the diagnosis of the financial result of housing and communal services enterprises]. Statistika i matematicheskie metody v jekonomike, № 3, S. 52–60. (in Russian)
Kargin B.B. (2017) Novye podhody k resheniju informacionnyh zadach v jekonomike i kommercii [New approaches to solving information problems in economics and commerce]. Theoretical and Practical Aspects of Economics and Intellectual Property, № 16, S. 235–240. (in Russian)
Maksimova O.M., Stjopina E.N. Ispol'zovanie nejronnyh setej dlja reshenija zadach stroitel'noj mehaniki [Using neural networks to solve structural mechanics problems]. URL: https://core.ac.uk/download/pdf/38644194.pdf/. (accessed: 14.07.2021). (in Russian)
Naumenko V.A. (2019) Primenenie nejronnyh setej dlja reshenija prakticheskih zadach v jekonomike [Application of neural networks for solving practical problems in economics]. Vektor jekonomiki, № 10, S. 1–12. (in Russian)
Grigor'eva D.R. Primenenie paketa statistica dlja analiza i prognozirovanija jekonomicheskih pokazatelej predprijatija [Application of the statistica package for the analysis and forecasting of the economic indicators of the enterprise]. URL: https://kpfu.ru/portal/docs/F2101409340/Grigoreva.pdf. (accessed: 10.08.2021). (in Russian)
