Problem Statement
In the highly competitive telecommunications industry, customer churn presents a significant challenge, directly influencing revenue generation and the long-term viability of companies. This project aims to uncover patterns and predictors and accurately forecast potential churn.
Data Exploration
Automatic payment methods like bank transfers & credit cards decrease customer churn, emphasizing the need for convenience and security. Early customer tenure is critical; a positive initial experience, competitive pricing, effective tech support, and incentives for long-term contracts enhance loyalty and reduce churn.
Model Evaluation
My analysis across GLM, Random Forest, KNN, and XGBoost, using oversampling, undersampling, and tuning, aimed to optimize accuracy, precision, recall, F1 score, and AUC. Random Forest and XGBoost excelled in accuracy and precision, while tuned KNN achieved high recall, effectively predicting customer churn.
Conclusion
Our analysis provides a strategic approach to understanding and mitigating customer churn through predictive analytics, offering insights for tailored retention strategies. Despite trade-offs between model precision and interpretability, our findings enable more informed decisions to enhance customer loyalty.
Toolkit
Programming Language: Julia;
Exploration: Gadfly, PlotlyJS, StatsPlots;
Models: CategoricalArrays, Distributions, GLM, DecisionTree, MLJ, MLJBase, MLJModels, MLJLinearModels, MLJGLMInterface, MLJDecisionTreeInterface, MLJXGBoostInterface, XGBoost, NearestNeighborModels;
Evaluation: FreqTables, HypothesisTests, StatsBase, Imbalance.