ДОСЛІДЖЕННЯ ЗАСТОСУВАННЯ НЕЙРОННИХ МЕРЕЖ ДЛЯ ЗАДАЧ ПЕРЕДБАЧЕННЯ ЗАБОРГОВАНОСТІ В КРЕДИТНІЙ СФЕРІ
RESEARCH OF NEURAL NETWORKS APPLICATION FOR DEBT PREDICTION TASKS IN THE CREDIT SPHERE
Сторінки: 52-57. Номер: №1, 2021 (293)
Автори:
В.В. РУСІНОВ
Національний технічний університет України “Київський політехнічний інститут імені Ігоря Сікорського”
V. RUSINOV
National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”
DOI: https://www.doi.org/10.31891/2307-5732-2021-293-1-52-57
Рецензія/Peer review : 08.01.2021 р.
Надрукована/Printed : 10.03.2021 р.
Анотація мовою оригіналу
В даній роботі розглянуто використання методів машинного навчання, а саме штучні нейронні мережі, для передбачення заборгованості в кредитній сфері. Розглянуті різні підходи до розв’язання суміжних задач на основі методів машинного навчання. Виконано розробку моделі передбачення заборгованості, проведено детальний аналіз запропонованої моделі, яка за рахунок покрокової зміни налаштування штучної нейронної мережі дозволяє покращувати розпізнавання потенційної заборгованості. Проведено експериментальні дослідження, застосовані різні оптимізатори та функції активації для досліджуваної мережі.
Ключові слова: машинне навчання, штучні нейронні мережі, кредитування.
Розширена анотація англійською мовою
This article researches the usage of machine learning methods, more specifically neural networks, for forecasting debts in credit sphere. Different approaches are researched that deal with similar tasks based on machine learning methods. The model of debt forecast is developed, analysis of this model is conducted, which uses step-by-step changes to the neural network allows to better detect potential debts. Research has been conducted using deep learning as a subset of machine learning for tasks that deal with debts in credit sphere and accounting. Articles show that there is potential in using neural networks for such tasks and they outperform standard algorithms such as linear regression.
Machine learning algorithms allow for creation of models that can detect changes in the data over a given period of time and using obtained from training process patterns to make predictions for new data. Such method will allow financial organizations to develop better strategies to avert unnecessary risks or to find ways to account for such risks.
During the process of model development, iterative approach has been taken to make closer steps to more accurately train on the supplied data. This is achieved through using old and tested algorithms, modern and more widely used and new emergent approaches to have a comparison between these approaches and to single out the best one. Using such approach we can see the how different optimizers and activation functions influence the overall model by using metrics such as accuracy and recall.
Using such approach, the full table of different approaches using various optimization algorithms and activation functions is provided to demonstrate the best model and how varied approaches change the overall accuracy of the model. This model can be used for further training on new data to adapt to a new economic situation.
Keywords: machine learning, artificial neural networks, finance and credit sphere.
References
- O. Partyn, O.V. Didukh. Features of the influence of the main factors on the financial stability of the company in terms of financial and economic crisis // Naukovyi visnyk natsionalnoho lisotekhnichnoho universytetu Ukrainy – 2010. – №10 – С.275-279
- Dreiseitl, S., & Ohno-Machado, L. (2002). Logistic regression and artificial neural network classification models: a methodology review. Journal of biomedical informatics, 35(5-6), 352-359.
- Santander Customer Transaction Prediction | Kaggle [Electronic resource] – Access to the resource: https://www.kaggle.com/c/santander-customer-transaction-prediction/data
- Bottou, L. (2012). Stochastic gradient descent tricks. In Neural networks: Tricks of the trade(pp. 421-436). Springer, Berlin, Heidelberg.
- Bae, K., Ryu, H., & Shin, H. (2019). Does Adam optimizer keep close to the optimal point?. arXiv preprint arXiv:1911.00289.
- Rieger, P., Gewald, H., & Schumacher, B. (2013). Cloud-computing in banking influential factors, benefits and risks from a decision maker’s perspective.
- Patidar, R., & Sharma, L. (2011). Credit card fraud detection using neural network. International Journal of Soft Computing and Engineering (IJSCE), 1(32-38).
- Gómez, J. A., Arévalo, J., Paredes, R., & Nin, J. (2018). End-to-end neural network architecture for fraud scoring in card payments. Pattern Recognition Letters, 105, 175-181.
- Bouzida, Y., & Cuppens, F. (2006, September). Neural networks vs. decision trees for intrusion detection. In IEEE/IST workshop on monitoring, attack detection and mitigation (MonAM) (Vol. 28, p. 29).
- Abyaneh, H. Z. (2014). Evaluation of multivariate linear regression and artificial neural networks in prediction of water quality parameters. Journal of Environmental Health Science and Engineering, 12(1), 1-8.
- Samanta, B., Al-Balushi, K. R., & Al-Araimi, S. A. (2006). Artificial neural networks and genetic algorithm for bearing fault detection. Soft Computing, 10(3), 264-271.
- Wang, C., Wang, Y., Ye, Z., Yan, L., Cai, W., & Pan, S. (2018, August). Credit card fraud detection based on whale algorithm optimized BP neural network. In 2018 13th International Conference on Computer Science & Education (ICCSE) (pp. 1-4). IEEE.
- Timothy Dozat. Incorporating Nesterov Momentum into Adam. ICLR Workshop, (1):2013–2016, 2016.
- Eger, S., Youssef, P., & Gurevych, I. (2019). Is it time to swish? Comparing deep learning activation functions across NLP tasks. arXiv preprint arXiv:1901.02671.