This is an end-to-end machine learning project that predicts whether a telecom customer will churn or stay. It works with the Telco Customer Churn dataset (7,043 rows, 20 features). The whole flow — cleaning data, training models, tuning hyperparameters, and saving the final model — is wrapped into one reusable pipeline.

1. Loaded and inspected the dataset with Pandas.
2. Split the data into 80% train and 20% test, keeping the churn ratio the same in both.
3. Built a preprocessing pipeline:
   • Numeric columns: median imputation + StandardScaler
   • Categorical columns: mode imputation + OneHotEncoder
   • Binary columns: passed through as-is
4. Combined preprocessing with two models: Logistic Regression and Random Forest.
5. Tuned both models with GridSearchCV (5-fold cross-validation).
6. Compared the models on accuracy, precision, recall, F1, and ROC-AUC.
7. Saved the best pipeline with Joblib so it can be reused with one line of code.

• Language: Python 3.11
• Data handling: Pandas, NumPy
• Machine learning: Scikit-learn (Pipeline, ColumnTransformer, GridSearchCV)
• Models: Logistic Regression, Random Forest
• Plots: Matplotlib, Seaborn
• Model saving: Joblib

• Logistic Regression won and was saved as the final model.
• Top churn signals: tenure, MonthlyCharges, and Contract type.
• Month-to-month customers churn the most.
• The saved pipeline handles missing values on its own, so it works on raw data with no extra steps.