Scholarlink

<p>Detecting credit card fraud is a complex challenge, requiring the development of a specialized anomaly detection framework and a model-explanation module. Existing methods rely primarily on data augmentation or standard machine learning models, whereas anomaly-detection-based approaches still need to be improved. Few studies have utilized AI interpretability methods to assess the significance of features in transaction data. This is vital for the black-box fraud detection module. We use our well-known dimensionality reduction statistical approach to incorporate XAI into anomaly detection algorithms like transformer, graph neural network (GNN), and transformer-GNN-ensembled. The study focuses on spotting inconsistencies in time series data, which is essential for detecting system flaws, fraud, or other unexpected behavior. The proposed technique is empirically validated on real-world time series datasets, demonstrating XAI's usefulness in improving forecasting accuracy and anomaly detection performance. This work also employs explainable artificial intelligence (XAI) to increase time series forecasting and anomaly detection accuracy. Time series data presents unique challenges for forecasting future trends and detecting aberrant patterns. Traditional forecasting approaches may need help to grasp the data's complexities and nonlinear relationships. The SHAP and LIME XAI techniques provide insights into the decision-making processes of complicated forecasting models, boosting their credibility and allowing for more accurate judgment. These approaches are used to examine the interpretability of these models, showing subtle differences and emphasizing trade-offs in stability, fidelity, and robustness. The study found that a transformer, GNN, and a transformer-GNN-ensembled model performed exceptionally well on Credit Card Fraud Detection and Credit Card Fraud Detection 2023, with the transformer achieving 99.93% accuracy, GNN 100% accuracy, and the transformer-GNN-ensembled model surpassing them all. The findings also show XAI's capacity to provide actionable insights, allowing stakeholders to make informed decisions based on a thorough understanding of the underlying 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

An efficient mechanism for time series forecasting and anomaly detection using explainable artificial intelligence

  • 期刊论文

  • 发表日期：

    2025年2月20日

查看全文
  • Amjad Iqbal,
  • Rashid Amin

摘要

Detecting credit card fraud is a complex challenge, requiring the development of a specialized anomaly detection framework and a model-explanation module. Existing methods rely primarily on data augmentation or standard machine learning models, whereas anomaly-detection-based approaches still need to be improved. Few studies have utilized AI interpretability methods to assess the significance of features in transaction data. This is vital for the black-box fraud detection module. We use our well-known dimensionality reduction statistical approach to incorporate XAI into anomaly detection algorithms like transformer, graph neural network (GNN), and transformer-GNN-ensembled. The study focuses on spotting inconsistencies in time series data, which is essential for detecting system flaws, fraud, or other unexpected behavior. The proposed technique is empirically validated on real-world time series datasets, demonstrating XAI's usefulness in improving forecasting accuracy and anomaly detection performance. This work also employs explainable artificial intelligence (XAI) to increase time series forecasting and anomaly detection accuracy. Time series data presents unique challenges for forecasting future trends and detecting aberrant patterns. Traditional forecasting approaches may need help to grasp the data's complexities and nonlinear relationships. The SHAP and LIME XAI techniques provide insights into the decision-making processes of complicated forecasting models, boosting their credibility and allowing for more accurate judgment. These approaches are used to examine the interpretability of these models, showing subtle differences and emphasizing trade-offs in stability, fidelity, and robustness. The study found that a transformer, GNN, and a transformer-GNN-ensembled model performed exceptionally well on Credit Card Fraud Detection and Credit Card Fraud Detection 2023, with the transformer achieving 99.93% accuracy, GNN 100% accuracy, and the transformer-GNN-ensembled model surpassing them all. The findings also show XAI's capacity to provide actionable insights, allowing stakeholders to make informed decisions based on a thorough understanding of the underlying models.