KERNEL FUZZY C-MEANS AND RNN-BASED FRAMEWORK FOR CUSTOMER CHURN ANALYSIS IN TELECOM

Abstract

A strategic method named customer relationship management (CRM) that builds profitable, offering long-lasting connections with important customers and other stakeholders. These contributions are considered to be the main objective of CRM. In the telecom sector, Customer churn is considered as one of the biggest CRM challenges. Then, maintaining current customers are far less expensive than finding new customers. This made churn as a major financial burden. For telecom sectors, the customer churn’s evolving problem has become a major concern. Getting novel customers is significantly more exclusive than holding current ones and it is revealed in this study. To address customer churn analysis and to improve the prediction models efficacy, sophisticated classification algorithms was employed in this study. Predicting individual customer’s churn behavior are the primary objective of the study. Effective data pre-processing is the main factor for this approach, as it includes the creation of derived features. Data comprehension is the initial step that supports in commercial significance.
In data pre-processing, limited data accessibility problem can be effectively addressed by implementing kernel-based fuzzy C-Means (KFCM) algorithm, as it converts the incomplete datasets into complete ones. Then, the churn analysis will support in detecting the signs of churn in highly valuable customers. Changes in customer status can be classified as: either partial defection (active use to suspension or non-use) or entire defection (active use to churn), as it will aid in differentiating the two. The mediating effects of partial defection on the connection among the churn determinants and total defection are examined by employing the Optimised XGBoost algorithm. This algorithm acts as a Base Classifier (BC). Here, complex interactions can be effectively handles by this BC, as it will also facilitate in detecting the non-linear patterns (NL) in the data. The Enhanced Gradient Boosting Machine (EGBM) has the ability in selecting the deriving features from customer profiles and call records.Proposed Recurrent Neural Network (PRNN)-based deep learning (DL) approaches like LSTM and GRU form the basis for churn prediction (CP). The standard models like Support Vector Machines (SVM), Logistic Regression (LR), Boosting Algorithms (BA), and Kernel Extreme Learning Machines (KELM) are compared with the proposed RNN model. Robust separation of churn data is offered and it was demonstrated by the experimental outcomes. It also offers high scalability and accuracy. The potential of DL techniques for addressing churn analysis problem in the telecom sector was demonstrated.