Scholarlink

<p>The recent trend in digital payments and debt-free lending of money in credit cards has spiked the increase of fraudulent credit card transactions. Many models were developed earlier but couldn’t perform better due to issues like data imbalance, private datasets, etc. Propose an ML-based credit card fraud detection approach using nature-inspired meta-heuristic algorithms for hyper parameter tuning. Extreme Gradient Boosting Classifier (XGBC), Categorical Boosting, and LightGBM are used as machine learning models. Elephant Herd Optimization (EHO), SMA and MSA are optimization algorithms. The proposed methodology performs a comparative analysis on 9&#xa0;ML and optimization algorithm combinations on a dataset extracted from the Paysim dataset. The data is balanced using the Borderline-SMOTE (synthetic minority oversampling technique). The comparative analysis demonstrates that the EHO + XGBC have an accuracy of 98% and a ROCAUC Score of 0.997, which outperformed the remaining models.</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

Optimizing Hyperparameters for Credit Card Fraud Detection with Nature-Inspired Metaheuristic Algorithms in Machine Learning

  • 期刊论文

  • 发表日期：

    2025年3月11日

查看全文
  • Shiva Shankar Reddy,
  • Kankanala Amrutha,
  • V. MNSSVKR Gupta,
  • K. VSSR Murthy,
  • V. Venkata Rama Maheswara Rao

摘要

The recent trend in digital payments and debt-free lending of money in credit cards has spiked the increase of fraudulent credit card transactions. Many models were developed earlier but couldn’t perform better due to issues like data imbalance, private datasets, etc. Propose an ML-based credit card fraud detection approach using nature-inspired meta-heuristic algorithms for hyper parameter tuning. Extreme Gradient Boosting Classifier (XGBC), Categorical Boosting, and LightGBM are used as machine learning models. Elephant Herd Optimization (EHO), SMA and MSA are optimization algorithms. The proposed methodology performs a comparative analysis on 9 ML and optimization algorithm combinations on a dataset extracted from the Paysim dataset. The data is balanced using the Borderline-SMOTE (synthetic minority oversampling technique). The comparative analysis demonstrates that the EHO + XGBC have an accuracy of 98% and a ROCAUC Score of 0.997, which outperformed the remaining models.