Re-activation of 17% churned bank clients with AI use in banking

One of the primary challenges faced was an imbalanced dataset, where only a small fraction of customers had churned. Therefore, it was crucial to ensure that the selected model accurately predicted this minority class with higher precision. The presence of such an imbalance could potentially lead to biased model performance. To address this issue, we conducted extensive research into existing solutions specifically designed for handling imbalanced data samples to mitigate any potential bias and improve the overall performance and accuracy of the model.

To evaluate the models’ precision, recall, and F-measure, we helped our client identify custom model metrics and acceptance criteria for each specific customer case in accordance with the business value. However, we have focused on F1-score as it illustrates a balance between precision and recall.

Our final solution encompassed a diverse range of machine learning algorithms, incorporating both classical boosting models and modern self-supervised techniques. By leveraging boosting models, we effectively addressed the original churn problem with a high degree of accuracy, ensuring precise predictions for customer churn.