DESIGNING OF THE SYSTEM FOR RANKING POTENTIAL CLIENTS OF A TELECOMMUNICATION COMPANY ON THE BASE OF MACHINE LEARNING
DOI:
https://doi.org/10.53920/ITS-2023-1-1Keywords:
call center, machine learning, learning by precedents, ranking system, business process analysis, classification algorithmAbstract
The work is devoted to the design of a system for ranking potential customers of a telecommunications company based on their probability of purchasing a product based on machine learning, which will allow optimizing the business process of working with "cold" customers.
This work examines the call center of a telecommunications company, aimed at making outgoing calls with the main purpose of selling the company's services. Despite the fact that different industries may have their own peculiarities in the activity of a call center, the general model of the organization remains approximately the same, which makes this study relevant for any industry.
It is proposed to use historical data to train a machine learning algorithm that will be able to emulate the activity of a supervisor in the formation of tasks for a call center.
For a better understanding of the current state of call center operation, the work presents a diagram of the business process of creating call tasks in BPMN notation. The paper also presents a diagram demonstrating the impact of the implementation of a system based on machine learning on the business process of creating tasks for callback.
This work investigates the ranking problem, which can be transformed into a binary classification problem. As part of the classification, it is necessary to determine the probability of potential customers belonging to one of two classes, which allows solving the problem of binary classification. The first class represents customers who are interested in the company's services, while the second class covers customers who do not show interest in the company's services. In this context, the most important thing is the probability that the potential customer belongs to the first class. After obtaining the probabilities of belonging to the first class, all potential customers are sorted in descending order of probability, thus solving the ranking problem.
