Predicting Customer Churn in Telecom Industry using MLP Neural Networks

Introduction Most telecommunications companies regard customers as the most important assets. Therefore, today's difficult problem for telecommunications companies is that customers leave the company to other service providers for some reason [1]. In most cases, this loss can occur at a rate that has a serious impact on the profitability of the company, as customers can easily change the company. In the market, the competition between telecommunications companies is rapidly expanding and the company has shifted focus from maintaining existing customers from acquiring new customers [1-3].

The reason why customers / customers quit company or change products is a big problem. In the wholesale telecommunications field, product loss plays a more important role than cancellation of customers. The neural network can predict customer confusion, which helps to understand the customer's next needs (such as content customization). It also helps to predict the best price range for economic change. - Identify the potential ways of employment and the timing to outsource installation / maintenance to contractors and improve the efficiency of supplier management. Because contractor employees may have a criminal record, it is beneficial to track large losses between contractors. Through the neural network you can predict which contract to use to improve security.

Neural networks are now widely used in various ways. From generation of image header to prediction of breast cancer, this diverse application is a natural result of various neural structures (feed forward neural network, convolution neural network etc). In all of these architectures, the special case of recurrent neural network Long-Term-Short-Term Memory (LSTM) is very successful in all aspects from machine translation, time series prediction, or general data to continuity doing. This is primarily due to the ability to memorize relatively long-term dependencies, which is achieved by considering further predictions from previous information.

An artificial neural network is just a deep neural network. There are other networks such as regression neural network (RNN), convolution neural network (CNN), Boltzmann machine. RNN can predict whether future stock price will go up or down. CNN is used in computer vision, identifies cats and dogs from a series of images, and identifies cancer cells present in the image of the brain. Boltzmann machine is used to program the recommendation system. Perhaps we can cover one of the neural networks of the future.