ABSTRACT
In recent days, telecommunications have been facing the critical issue of predicting the customer churn that can help in maintaining customers. It is far more expensive to acquire new customers than to retain existing ones. For that reason, the major carriers are trying to develop models to anticipate that there are a lot of customers and to take appropriate action. The majority of industries that are dynamic in diligence and have low switching costs are concerned with customer churn in the telecom sector. With an approximate monthly churn rate of 30, telecommunications assistance ranks highest among all businesses affected by this problem. There are a variety of methods for dealing with such a problem, including the creation of predictive systems that are difficult to trace and define. Therefore, the construction of any such model would be highly complex, involving the use of Machine Learning. Initially, the model was created using the Decision Tree algorithm in the model structure phase. Later, after comparing the results of various Machine Learning algorithms, a Random Forest classifier, and Ada Boost, XGBoost, and Decision Tree algorithms were used to create models for churn analysis. The upgraded decision tree set automatic learning technique utilized for categorization in this suggested system is called Improved Gradient Boosted Decision Trees (I-GBDT). The objective is to get a data-driven resolution which will permit us to scale back churn rates and, as a consequence, to extend client satisfaction and corporation revenue. The decision tree, which is known for its effectiveness, is utilized in the development of a churn analysis model to analyse the behavior of telecom customers. Through this improved version of decision tree analytics, 93% accuracy is achieved. In addition, this approach helps telecom customers to make them profitable. The use of alternative strategies for dealing with data imbalance and testing their applicability has been suggested by additional research.
