Comparison of Random Survival Forest and Fuzzy Random Survival Forest Models in Telecommunications Industry Customer Data

Abstract

The telecommunications sector is facing increasing competition, and customer churn is still a major
challenge despite the implementation of advanced promotions and high-quality services. Churn refers to
the discontinuation of services by customers, influenced by several factors that can be found through data
modeling. This study compares two predictive models, Random Survival Forest (RSF) and Fuzzy Random
Survival Forest (FRSF), for predicting customer churn time in the telecommunications industry. Both
models are evaluated using the median C-index value obtained from 20 repetitions, ensuring more
consistent and reliable results. RSF, a widely used survival analysis method, has shown strong predictive
power, with studies reporting up to 99% accuracy in churn prediction. However, FRSF, a modified version
that incorporates fuzzy logic, has proved superior performance, particularly in handling imprecise or
uncertain data. The results show that FRSF achieves a lower error rate of 0.1739, compared to RSF's error
rate of 0.1906. These findings suggest that FRSF outperforms RSF in churn prediction, making it a more
reliable and righter model for finding at-risk customers. The study concludes that the FRSF model is the
preferred choice for predicting churn in the telecommunications industry, offering better predictive quality
and consistency in handling uncertain data.