Abstract:
"In the rapidly changing e-commerce landscape, businesses are always trying to improve customer retention and mitigate churn. Customers leaving a company is simply known as customer churn, and it is a significant threat. Traditional churn prediction models, while useful, frequently fail to offer the insights needed for efficient risk level analysis. In order to fill this gap, this study introduces an advanced churn prediction framework that uses explainable AI (XAI) principles in combination with a group of machine learning models to analyse customer churn risk at a comprehensive level.
In order to address this complex issue, the study first employs the individual performance of a wide range of basic machine learning models, namely XGBoost, Random Forest, Logistic Regression, LightGBM, and K-Nearest Neighbors, in predicting customer turn over. The research innovates by using Stacking Ensemble Method, which combines individual models to improve overall prediction accuracy, recognizing the limitations of relying on single model predictions. Additionally, the study incorporates XAI techniques to address the need for transparency and interpretability in AI-driven decision-making process.
The use of the ensemble method has demonstrated promising initial outcomes. The author evaluated the model’s performance by using Confusion Matrices and ROC/AUC metrics. With an AUC score of 0.98, the Stacking Ensemble method validates its effectiveness to improve prediction accuracy. In contrast, the AUC score of other used models is comparatively lower, highlighting the advantage of the proposed ensemble and XAI approach. These preliminary findings validate the effectiveness of combining multiple predictive models but also highlight the importance of explainability."
