Customer Churn Analysis

A supervised learning project that analyzes customer behavior data and demographic information to identify customers at risk of discontinuing service with over 90% accuracy.

This study involves a comprehensive data mining analysis on anonymized large datasets belonging to the telecommunications sector. The main objective of the project is to create a decision support system that enables the development of proactive strategies to minimize customer churn. Project Process and Methodology: Data Preprocessing: Completing missing values ​​of categorical variables, outlier analysis, and one-hot encoding were performed. Exploratory Data Analysis (EDA): Correlations and distributions in the dataset were visualized using Matplotlib and Seaborn libraries; the most important features affecting customer churn were identified. Model Development: Models were trained using the Scikit-learn library with Logistic Regression, Random Forest, Decision Trees, and XGBoost algorithms. Performance Evaluation: The success of the models was compared using Confusion Matrix, F1-Score, Accuracy, and ROC-AUC curves. The model was optimized using the highest-performing Random Forest algorithm (Hyperparameter Tuning).