Abstract:
Customer churn is one of the most critical issues in telecom companies. As it directly affects the company’s revenue, arises the need of finding ways to predict and then prevent this kind of phenomen. Machine learning can highly contribute in developing algorithms that can be used in various companies that can firstly indicate factors affecting and then create patterns. The study aims to emphasize and investigate a conjecturing analysis of most of the predictive algorithms used for customer churn prediction, in telecommunication. Including the key factors affecting this kind of customer behavior, the causes and the consequences for the companies and concluding with how it can be predicted using machine learning algorithms, giving a hand to the companies to take measures before they experience their customer loss. We will use a real dataset obtained by an Albanian telecom company, Vodafone. The algorithm tested will be Logistic Regression and Random Forest. Models will be compared accordind to some evaluation metrics and the outperforming model will our suggestion to the company.The study brings in focus the useful indications for telecommunication companies and suggests some marketing strategies that use algorithmic outcomes to reduce churn rates. Within both of the algorithms Random Forest showed an outstanding performance with an accuracy of 94%, while the Logistic Regression struggled at an accuracy level of 85%. These results indicated that Random Forest is a better practice for this classification. As the dataset obtained consists of mainly categorical data, it is easier for Random Forest to deal with it, while Logistic Regression struggles when it comes to categorical data. These results will serve as a helpful insight for telecom companies to face the issue of customer churn.
