Scholarlink

<p>The increasing volume of financial transactions has significantly amplified the risk of credit card fraud, demanding data-driven and reliable detection mechanisms. This study aims to develop an intelligent fraud detection framework that integrates Multi-Objective Harris Hawks Optimization (MO-HHO) with K-means SMOTE to address severe class imbalance while jointly optimizing feature relevance and redundancy. The proposed multi-objective strategy balances exploration and exploitation by optimizing Pearson correlation, mutual information, and chi-square metrics, enabling the selection of highly discriminative and minimally redundant features. The optimized feature subset is evaluated using a Random Forest classifier, resulting in improved predictive accuracy and robustness. Experimental validation conducted on benchmark datasets, including the European Credit Card (ECC) dataset and the PaySim dataset, demonstrates accuracy levels of 99.13% and 98.10%, respectively. The framework is further evaluated under different sampling strategies, including no balancing, SMOTE, and K-means SMOTE with varying ratios (1:1 and 80:20), to assess robustness under realistic data distributions. The results highlight the effectiveness of the proposed framework in detecting fraudulent transactions while maintaining practical applicability in real-world financial systems.</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

Credit card fraud detection using multi-objective harris hawks optimization and K-means SMOTE

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Mosleh M. Abualhaj,
  • Raman Kumar,
  • Sarbeswara Hota,
  • Mahendihasan S. Heera,
  • Gowrishankar J,
  • Sushant Jhingran,
  • Ankayarkanni B,
  • Sumit Sharma,
  • Anita Gehlot

摘要

The increasing volume of financial transactions has significantly amplified the risk of credit card fraud, demanding data-driven and reliable detection mechanisms. This study aims to develop an intelligent fraud detection framework that integrates Multi-Objective Harris Hawks Optimization (MO-HHO) with K-means SMOTE to address severe class imbalance while jointly optimizing feature relevance and redundancy. The proposed multi-objective strategy balances exploration and exploitation by optimizing Pearson correlation, mutual information, and chi-square metrics, enabling the selection of highly discriminative and minimally redundant features. The optimized feature subset is evaluated using a Random Forest classifier, resulting in improved predictive accuracy and robustness. Experimental validation conducted on benchmark datasets, including the European Credit Card (ECC) dataset and the PaySim dataset, demonstrates accuracy levels of 99.13% and 98.10%, respectively. The framework is further evaluated under different sampling strategies, including no balancing, SMOTE, and K-means SMOTE with varying ratios (1:1 and 80:20), to assess robustness under realistic data distributions. The results highlight the effectiveness of the proposed framework in detecting fraudulent transactions while maintaining practical applicability in real-world financial systems.