• Conducted EDA showing the distribution and feature correlation based on the dataset with 7000+ clients and 21 columns, utilized Cohort Analysis (Seaborn) to segment customers based on tenure feature which is selected by feature importance plot.
• Trained and tested 3 predictive tree-based models (Random Forest, AdaBoost, Gradient Boosting) to evaluate the best performance model; Compiled the best model AdaBoost with the best weighted average f1-score of 83%, 82% precision, and 704 support.
The Original dataset is from:
https://www.kaggle.com/datasets/blastchar/telco-customer-churn