Created an interpretable model that predicts churn on telecom data found on Kaggle (F1: 0.796) to build a pipeline to try and retain more customers. Used SMOTE to oversample minority class (churned customers) which improved the F1 score from 0.61 to 0.80. Engineered new features, for example, aggregating different features together. Performed feature selection using SHAP values. Optimized Logistic Regression and Decision Tree models using Optuna and pruned the decision tree further to reduce overfitting. Used Mlflow to track model experimentation.