Machine Learning-based Customer Churn Analysis in Telecommunications Using Support Vector Machines

Faced with globalization and increasing competition, the information available via the Internet and the many connected objects continues to increase. This explosion of data, often heterogeneous and from diverse sources, poses major challenges in terms of storage, analysis and exploitation. This paper is the result of the present research on the analysis and classification of churning customers in a telecommunications company. These data, often heterogeneous and coming from various sources, require in-depth analysis as well as new storage and exploration paradigms to extract value from them. The dataset used for the implementation of the prediction model is based on the existing reality, within the telecommunications company named Airtel Congo; on the customer management policy, more precisely the customers who are candidates for churn. In the telecommunications sector, companies accumulate large amounts of information about their customers, coming from multiple sources: social networks, telephone platforms, electronic messaging, open data, geolocation, and many others. The intelligent exploitation of this data allows to better understand user behavior and anticipate key phenomena, such as “ churn ” – i.e. customer unsubscription. Churn is a major strategic issue for telecommunications companies, as customer loss leads to high costs related to new subscriber acquisition and reduced revenue. Thus, identifying customers at risk of churn and understanding the underlying factors are essential to implement preventive actions and build customer loyalty. In this study, a machine learning model based on support vector machines (SVM) was proposed to analyze and classify churning customers. This algorithm, recognized for its ability to handle complex and multidimensional data, is implemented using the LIBSVM library in the C# language. The objective is to build a powerful predictive model to identify, with high accuracy, customers likely to leave the operator, in order to optimize retention strategies and maximize customer satisfaction. Based on various techniques such as supervised and unsupervised learning, it allows to discover hidden patterns and make accurate predictions. SVM, in particular, illustrate the effectiveness of supervised approaches, by allowing an optimal separation of classes through the maximization of the margin.