Analisis Fitur Dan Prediksi Loyalitas Pelanggan Menggunakan Algoritma Pembelajaran Mesin

The telecommunications industry faces intense competition, making it crucial to understand customer behavior in order to reduce customer churn a condition where customers discontinue their subscriptions. This study aims to identify the most influential features affecting churn and to build a predictive model using two machine learning algorithms: XGBoost and Random Forest. The dataset used in this research is the Telco Customer Churn dataset from Kaggle, consist-ing of 7,043 customer records. The research process includes data preprocessing, feature selection using Sequential Feature Selection (SFS), and model interpreta-tion with SHAP (Shapley Additive Explanations) to explain the contribution of each feature to churn prediction.
The results show that the features Contract, Tenure, and MonthlyCharges are the most influential factors in predicting churn. The application of SFS improved the model performance, with Random Forest achieving an accuracy of 0.7915, while XGBoost obtained 0.7863. The SHAP analysis reinforced these findings by re-vealing that customers with long-term contracts and longer subscription periods tend to have a lower likelihood of churn. The combination of SFS and SHAP not only enhances model accuracy but also provides deeper interpretability regard-ing the key factors influencing customer churn.