Research of machine learning methods and their application for forecasting use outflow by telecommunications services

DOI: 10.31673/2412-9070.2020.042231

Abstract

The article examines the problem of outflow of customers of telecommunications companies: foreign and Ukrainian. It is determined that it is appropriate for companies to use information systems that will predict the behavior of users of services, and will warn the company in the presence of risks. It is established that foreign companies are more progressive in research and customer retention. In Ukrainian companies, unstructured data research tools are outdated. In order to expand the tools of Ukrainian companies, the achievements of foreign scientists have been studied. When processing a large amount of information in order to build short- and medium-term forecasts of consumer behavior, it is appropriate to build models of user behavior and use machine learning methods. It is established that scientists prefer the method of random forest. The methods of machine learning are analyzed in the work: their advantages and disadvantages are determined. The characteristics of the random forest method are investigated. It is established that further improvements in this area may be compositions of algorithms. The method of composition of algorithms — boosting is investigated. Its advantages and disadvantages are established, features of this method are defined. The stages of building a model for creating a forecast are defined. Based on the analyzed literature on the outflow of users, proposals have been identified that can reduce the reduction in their number. The priority directions of further researches concerning optimization problems of methods of machine learning, in particular in the conditions of uncertain factors are established.

Keywords: outflow of users; modeling; machine learning methods; boosting; unstructured data.

References
1. Condamoor Ravi Building Predictive Models for Customer Churn in Telecom. URL: https://www.experfy.com/blog/building-predictive-models-for-customer-churn-in-telecom.
2. Canale A., Lunardon N. Churn prediction in telecommunications industry. A study based on bagging classifiers telecom // Carlo Alberto Notebooks, 2014. Vol. 350. P. 1–11. URL: https://www.carloalberto.org/assets/working-papers/no.350.pdf.
3. Khan A. A., Sanjay J., Sepehri M. M. Applying data mining to customer churn prediction in an Internet
service provider // Int. J. Comput. Appl. 2010. Vol. 9, No. 7. P. 8–14. URL: http:// www.ijcaonline.org/volume9/number7/pxc3871889.pdf.
4. Telecom Churn Analysis. URL: https://towardsdatascience.com/customer-churn-analysis-4f77cc70b3bd
5. Neal A. Akyildirim Brief Overview of Customer Churn Analysis and Prediction with Decision Tree Classifier. URL: http://rstudio-pubs-static.s3.amazonaws.com/277278_427ca6a7ce7c4eb688506efc7a6c2435.html
6. Скоринговое моделирование финансовых потоков от взыскания / Т. И. Григорчук, З. В. Максименко, Л. Ф. Розанова, Г. Р. Бикбулатова // Нефтегазовое дело: електрон. версія журн. 2015. № 5. С. 630–655.
7. Воронцов. Курс «Машинное обучение» 2019 (Школа анализа данных). URL: https://ya-r.ru/2020/05/07/vorontsov-kurs-mashinnoe-obuchenie-2019-shkola-analiza-dannyh/