Predictive modeling of motor client's propensity to churn

This master's thesis focuses on the predictive modeling of motor client's propensity to churn at Lusitania Seguros company in Portugal, Utilizing a dataset of 248,218 observations with 45 distinct features from 2021 to 2022, the study highlights the significance of understanding customer churn in the competitive insurance industry. The research methodology involves applying various machine learning models including Logistic Regression, Decision Tree, Random Forest, XGBoost, K-Nearest Neighbors, and LightGBM. A significant part of the study was the use of the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance, coupled with rigorous validation methods like stratified K-fold cross-validation and a hold-out test set. The findings indicate a close performance between the LightGBM and XGBoost models. LightGBM slightly outperforms with an accuracy of 0.98, demonstrating high recall, precision, and F1-score for the minority class. XGBoost closely matches these metrics, presenting itself as an equally effective alternative for churn prediction. The thesis also delves into feature importance analysis from the LightGBM model, providing valuable insights for targeted customer retention strategies at Lusitania Seguros.