Identifying ISP’s customers profiles prone to Churn by employing TR-069 protocol data into ML algorithms

Authors

  • Bernardo Gatto Universidade Federal de Santa Maria (UFSM) - Campus Frederico Westphalen
  • Patricia Lengert Universidade Federal de Santa Maria (UFSM) - Campus Frederico Westphalen
  • Augusto Abel Instituto Federal de Educação, Ciência e Tecnologia Farroupilha (IFFar) - Campus Frederico Westphalen
  • Yuri Bandeira Instituto Federal de Educação, Ciência e Tecnologia Farroupilha (IFFar) - Campus Frederico Westphalen
  • Raul Ceretta Nunes Universidade Federal de Santa Maria (UFSM)
  • Ricardo Tombesi Macedo Universidade Federal de Santa Maria (UFSM) - Campus Frederico Westphalen

DOI:

https://doi.org/10.5335/rbca.v16i2.15312

Keywords:

Customers Profiles, Internet Service Provider, TR-069 Protocol

Abstract

Internet Service Providers (ISPs) offer an essential communication infrastructure to support people everyday's tasks over the Internet and also the advent of new networking technologies. However, a challenge to ISPs is to reduce churn rate, also known as low customer retention rate. Despite efforts in the literature, ISPs remain short of tools to identify customers' churns. This paper proposes ChurnSense, a process to identify ISP customers profiles by employing machine learning techniques, contributing to the understanding of the churn problem. The processes comprises three steps: Collect, Pre-processing, and Analysis. Through it, Collect gathers data from TR-069 protocol, Pre-processing treats these data and Analysis finds clusters that define customers profiles, providing useful information to decision making about churn. A case study was conducted by employing real data from a regional ISP. The obtained results show 20.61% of customers devices with connection quality below the expected, being at risk of churn.

Downloads

Download data is not yet available.

Published

2024-08-05

Issue

Section

Original Paper

How to Cite

[1]
2024. Identifying ISP’s customers profiles prone to Churn by employing TR-069 protocol data into ML algorithms. Brazilian Journal of Applied Computing. 16, 2 (Aug. 2024), 16–30. DOI:https://doi.org/10.5335/rbca.v16i2.15312.