Customer Churn Prediction in the Telecom Sector with Machine Learning and Adaptive k-Means Cluster using Imbalance Data

A crucial aspect of maintaining a customer-oriented business in the telecommunications sector with machine learning (ML) is understanding the reasons and factors that lead to customer churn. However, the dataset is difficult by noise, misclassifications, duplicated data, and imbalanced customer information complicating the process of identifying ways to split the dataset based on events. The under-sampling method address the imbalance data by decreasing the data quantity of majority class, thereby achieving a balanced dataset. This method, utilizing Adaptive k-means clustering, directly determines the reduction based on that data quality in each class. The datasets are combined with class labels to generate a new balanced dataset used for classification in the SVM algorithm, which distinguishes between customer churn and non-churn in the prediction for the telecom sector. The SVM algorithm can perform linear and non-linear classification with a kernel function to control overfitting improve generalization and handle high-dimensional data. The obtained results demonstrate that the proposed method achieves better accuracy of 95.70%, precision of 56.01%, Recall of 55.05%, and F1-score of 96.05% on the telecom churn dataset. These results ensure superior detection and classification performance compared to other existing methods, such as Random Forest (RF), Decision Tree (DT) and K-Nearest Neighbour (KNN).