Semi-supervised Curriculum Ensemble Learning for Financial Precision Marketing

This paper tackles precision marketing in financial technology, focusing on the accurate prediction of potential customers' interest in specific financial products amidst extreme class imbalance and a significant volume of unlabeled data. We propose the innovative Semi-supervised Curriculum Ensemble (SSCE) framework, which integrates curriculum pseudo-labeling and balanced bagging with tree-based models. This novel approach enables the effective utilization of high-confidence predicted instances from unlabeled data and mitigates the impact of extreme class imbalance. Experiments conducted on a large-scale real-world banking dataset, featuring five financial products, demonstrate that the SSCE consistently outperforms existing methods, thereby promising significant advances in the domain of financial precision marketing.