not-yet-known not-yet-known not-yet-known unknown In the fast-growing and competitive telecommunications business, customer churn is a major problem since it directly affects the company’s revenue. Therefore, it is crucial to identify the factors that influence customer churn. Traditional Feature selection has yielded fewer effective results. Few studies have used simulated annealing for feature selection and classification for churn prediction in Telkom. Most of these studies used fixed cooling schedules after each iteration of the process. Very Slow cooling demands a substantial amount of computation time and, therefore can be impractical for large or complex problems where faster convergence is desired. Fast cooling can lead to premature convergence and getting trapped in local minima due to faster cooling rates. This results in the system settling into a local minimum, which is a solution that is better than its immediate neighbors but not the best overall. To strike a balance between efficiency and computational time, this study proposes developing an adapted simulated annealing approach. This method dynamically adjusts the cooling rate based on feedback on the progress of solution quality of the search process, ensuring adequate exploration while progressively focusing on convergence. In this study, we use accuracy as the solution quality of algorithms. We conducted experiments using machine learning algorithms like Random Forest, Decision Trees, XGBoost and Adaboost blended with Adapted simulated annealing (ASA) using publicly available telecommunications datasets. Results show our ASA method outperforms traditional SA cooling schedules and other metaheuristic algorithms in accuracy, precision, ROC, recall, and F1-score.