Scholarlink

<p>As business models shift from product-centric to customer-centric, financial institutions have transitioned from passively responding to micro and small enterprise (MSE) loan requests to proactively acquiring customers. Consequently, customer demand prediction (CDP) has become an indispensable component of credit decision-making. Dynamic selective ensemble learning, with its ability to capture diverse and complex demand patterns, holds great potential in improving CDP performance. However, key challenges remain, particularly in adapting to local nonlinearity and misclassification independence among base classifiers in complexity data distributions. To this end, we propose a bilevel dynamic selective ensemble learning method for CDP in MSE loan products. Specifically, we design a mutual information-based instance selection module to accommodate complexity data distribution with local nonlinearity adaptation. We also propose a double fault-based classifier selection module to select base classifiers with high complementarity and misclassification independence. Empirical evaluation demonstrates that the proposed method outperforms all state-of-the-art benchmarks. This study provides financial institutions with a novel tool for proactively identifying MSEs with loan demand, thereby enhancing the efficiency of loan product reach and significantly reducing customer acquisition costs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting Customer Demand for MSE Loan Products: A Bilevel Dynamic Selective Ensemble Learning Method

  • 期刊论文

  • 发表日期：

    2025年11月5日

查看全文
  • Pingfan Xia,
  • Zhao Wang,
  • Xuhui Zhu

摘要

As business models shift from product-centric to customer-centric, financial institutions have transitioned from passively responding to micro and small enterprise (MSE) loan requests to proactively acquiring customers. Consequently, customer demand prediction (CDP) has become an indispensable component of credit decision-making. Dynamic selective ensemble learning, with its ability to capture diverse and complex demand patterns, holds great potential in improving CDP performance. However, key challenges remain, particularly in adapting to local nonlinearity and misclassification independence among base classifiers in complexity data distributions. To this end, we propose a bilevel dynamic selective ensemble learning method for CDP in MSE loan products. Specifically, we design a mutual information-based instance selection module to accommodate complexity data distribution with local nonlinearity adaptation. We also propose a double fault-based classifier selection module to select base classifiers with high complementarity and misclassification independence. Empirical evaluation demonstrates that the proposed method outperforms all state-of-the-art benchmarks. This study provides financial institutions with a novel tool for proactively identifying MSEs with loan demand, thereby enhancing the efficiency of loan product reach and significantly reducing customer acquisition costs.