Customer churn prediction model for E-commerce using an enhanced adaptive boosting algorithm

Customer attrition rates are still high as the e-commerce sector becomes more competitive. To increase client retention, churn risk must be accurately predicted. However, when it comes to high-dimensional sparse and unstructured data, typical customer churn prediction models perform poorly, and it is challenging to adjust to the dynamic shifts in customer behavior and issues with category imbalance. The study suggests an e-commerce customer churn prediction model based on the enhanced adaptive boosting technique of multilayer perceptrons in order to address the aforementioned issues. Firstly, the study optimizes the traditional adaptive boosting framework and introduces the misclassification penalty coefficient in the loss function. Then, it combines the principal component analysis dimensionality reduction technique to effectively handle high-dimensional sparse data and improve the computational efficiency. The experimental results show that the N-AdaBoost model achieved accuracy rates of 97.61% and 97.81% on the two datasets, with recall rates of 92.68% and 91.99%, and F1 scores of 0.9947 and 0.9924, respectively. These values are superior to those of the comparison models. The ablation experiments demonstrated that removing the SMOTE module led to a 9.52 percentage point decrease in recall rate, and removing the dynamic weight mechanism resulted in a 6.75 percentage point decrease in recall rate, verifying the effectiveness of each core module. This study provides e-commerce enterprises with an accurate customer churn prediction tool to support the formulation of refined operation and customer retention strategies, which is of great practical significance.