Source-linked AI summary
A Neural Network based Approach for Predicting Customer Churn in Cellular Network Services
Anuj Sharma, Dr. Prabin Kumar Panigrahi
TL;DR
Customer churn prediction matters because churn can reduce business and retention efforts depend on identifying likely churners. The paper applies a neural-network model to cellular-service customer data and reports over 92% overall accuracy, while detecting true churners less reliably than loyal customers. The authors identify preprocessing, hybrid methods, and testing in other sectors as future directions.
Problem
Accurate churn prediction is needed to help cellular operators manage customer retention as competition and the business impact of churn increase.
Method
The paper implements a neural network in Clementine using cellular-service customer data and supervised training with a reserved test set.
Results
Over 92% overall accuracy was reported; the model correctly predicted over 97% of loyal customers but only 66% of true churners.
Takeaways & Limitations
The results suggest data-mining techniques can support an early-warning model for customer churn management.
Takeaways & Limitations
The methodology remains to be tested in other sectors and compared with other prediction techniques, with preprocessing and feature selection identified as future work.
Abstract
from arXiv · showhide
Marketing literature states that it is more costly to engage a new customer than to retain an existing loyal customer. Churn prediction models are developed by academics and practitioners to effectively manage and control customer churn in order to retain existing customers. As churn management is an important activity for companies to retain loyal customers, the ability to correctly predict customer churn is necessary. As the cellular network services market becoming more competitive, customer churn management has become a crucial task for mobile communication operators. This paper proposes a neural network based approach to predict customer churn in subscription of cellular wireless services. The results of experiments indicate that neural network based approach can predict customer churn.
1. INTRODUCTION
Customer churn threatens revenue, making accurate prediction important for targeted retention. This paper applies an artificial neural network to predict churn in cellular wireless services.
- Customer churn can reduce business and profit, while small retention-rate changes may have significant business impact.
- Effective churn management requires accurate models that discover patterns and predict customer behavior from available data.
- Neural networks, support vector machines, and logistic regression are among the techniques used to predict customer churn.
- Churn management includes predicting likely churners and assessing how operators should respond through retention actions.
- The paper supports churn prediction by proposing an artificial neural network model based on stored customer data for cellular operators.
2. LITERATURE REVIEW
The literature frames churn as customer movement between providers and reviews predictive and retention-management approaches. This paper positions a neural-network approach within that research and cellular-service context.
- 2.1 Customer Churn Management: Customer churn denotes movement from one cellular provider to another, while churn management concerns retaining profitable customers.
- 2.1 Customer Churn Management: Operators may prioritize retention using customer profitability, churn propensity, or both, but profitability measures remain nonstandardized in developing-country telecom markets.
- 2.1 Customer Churn Management: Reactive retention waits for cancellation requests, whereas proactive retention identifies likely churners before cancellation and may offer incentives.
- 2.1 Customer Churn Management: Inaccurate churn predictions can waste incentive spending on customers who would not otherwise churn.
- 2.2 Related Work: Prior studies used support vector machines, logistic regression, Markov chains, random forests, clustering, and neural networks across several industries.
- 2.2 Related Work: This paper proposes a neural-network approach for predicting churn among cellular wireless-service subscribers.
3. THE PROPOSED NEURAL NETWORK (NN) BASED APPROACH
The proposed approach uses a multilayer neural network trained by forward and backward propagation on cellular-service customer data. Implementation in Clementine uses customer variables, supervised training, and a reserved test set.
- 3.1 Artificial Neural Network: Artificial neural networks use interconnected processing nodes and nonlinear transfer functions to map inputs to estimated outputs.
- 3.1 Artificial Neural Network: A multilayer perceptron contains input, hidden, and output layers, with weighted connections between successive layers.
- 3.1 Artificial Neural Network: Back-propagation training computes outputs, measures errors against desired outputs, and updates connection weights backward through the network.
- 3.2 Dataset: The dataset contains cellular-service customer records assembled from provider databases, including billing and customer-service information.
- 3.2 Dataset: Inputs include account, plan, usage, calling, charge, and customer-service variables, while state and phone number serve as identification fields.
- 3.3 Training and Testing of Neural Network: Clementine’s neural-network implementation uses input variables, hidden neurons, and an output target, with the Quick method selecting a feed-forward back-propagation topology.
- 3.3 Training and Testing of Neural Network: A randomly selected portion of the data trains the network, while the remaining data are reserved to test generalization and performance.
4. RESULT AND DISCUSSION
The neural-network model achieved high overall test accuracy and identified input variables associated with potential churn. Its predictions were substantially better for loyal customers than for true churners, making the model more suitable for retaining customers than detecting every loss-causing churner.
- 92.35% test-set accuracy was achieved by the neural network, whose best configuration used one hidden layer with three neurons.State and phone number were excluded because they served only as identifiers.
- Frequent calls to customer service numbers, an international calling plan, and high daytime calling charges were associated with potential churners.
- The generated network produced predicted churn labels and confidence values for each input record, then compared predictions with actual churn values.The confidence value ranges from 0.0 to 1.0, with higher values indicating greater confidence.
- Over 97% of loyal customers were classified correctly, compared with only 66% of true churners.
- The model is therefore more reasonable for identifying customers who will not churn than for identifying customers whose churn would cause company losses.
5. CONCLUSION AND FUTURE RESEARCH
Churn prediction is presented as an important problem for competitive cellular operators seeking proactive customer retention. The paper concludes that data mining can provide an early-warning model with more than 92% overall accuracy, while proposing broader validation and model extensions for future research.
- Competitive cellular operators need accurate churn prediction to identify possible churners and take proactive actions to retain valuable loyal customers.
- More than 92% overall accuracy was reported for predicting customer churn with the final model.
- Future work includes dimensionality reduction or feature selection, hybrid models combining neural networks with other techniques, and testing in sectors beyond telecommunications.