Feature Selection for Customer Churn Prediction: A Review on the Methods & Techniques applied in the Telecom Industry

Abstract

The rapid growth of the telecommunication industry has significantly contributed to the rise of interest in customer churn prediction studies within the telecom domain. Customer retention is one of the significant factors affecting the livelihood of businesses across every industry. Therefore, it is many companies’ priority to safeguard the existing customers, from cancelling their subscription and migrating to the competitors’ company, which is also known as customer churn. One of the most crucial processes in achieving a high-performance churn prediction model is feature selection. Unfortunately, this process is not highlighted in many studies within this domain. It was found that few studies did not even include the feature selection process as part of their data preparation from several articles published between 2016 and 2021. Countable studies collected, were conducted evolving around feature selection by comparing different methods. Moreover, the researchers whose data preparation process involved feature selection used manual feature selection and automatic feature selection. The automatic feature selection can then be further classified into filter method, wrapper method and embedded method. The most prominent feature selection method found in this survey was Fisher Score, which is a filter method, followed by Random Forest, which is an embedded method.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 9 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.