| Summary: | As the market becomes gradually more saturated and competition is fierce among the telecomunication companies, keeping and retaining existing customers become a priority to service providers. Customer retention program is one of the main strategies adopted in order to keep customers loyal to their provider. However, it requires a high cost and therefore the best strategy that companies could practice is to focus on identifying which customers have the potential to churn at an early stage. The targetted approach should be applied because it does not only lower cost, but the impact is very high to ensure the profitability of the company. Prediction of cutomer's potential to churn is not a straight forward prediction without any systematic mechanism or tools to predict. Common churn management process focuses on building a predictive model using past churn data and identifying the factors of customer churn. Motivated by the limited amount of research on investigating customers churn using machine learning techniques, this research explored the potential of an artficial neural network to improve customer churn prediction. The dataset used to train and test the neural network provided by one of the leading telecomunication company in Malaysia. This research proposes Multilayer Perceptron (MLP) neural network approach to predict customer churn based on 14 proposed predictors. The neural network models are trained respectively by nine (9) backpropagation (BP) techniques. The results are compared against popular churn prediction techniques such as Multiple Regression Analysis and Logistics Regression Analysis. The result comfirmed previous claims made by many researchers stating the superiority of neural over statistical models in prediction tasks. The best results of the experiment indicate that this model able to produce clasification accuracy of 94.82%. Overall, the findings suggest that a neural network in modelling techniques offers a viable alternative to traditional predictive approahes in customers churn prediction.
|
|---|