引导式测验 · 中等 等级

人工智能在欺诈检测测验中的应用

审查欺诈模型错误警报、链接网络信号、速度、测量漂移、FinCEN 深度伪造警报报告和调查员审查。

相关引导路径人工智能在欺诈检测中的应用
问题 1 的 8

人工智能解决的纯粹基于规则的欺诈系统的主要局限性是什么？

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. 人工智能解决的纯粹基于规则的欺诈系统的主要局限性是什么？
    • 犯罪分子学习固定规则并绕过它们
    • 它们运行成本太高
    • 他们无法处理数字
    • 他们只处理现金交易
  2. 为什么误报是欺诈检测中的一个严重问题？
    • 他们减慢了服务器的速度
    • 他们合法地阻止真正的客户并削弱信任
    • 他们无需花费任何费用即可修复
    • 它们只会影响欺诈者
  3. 哪种技术对于揭露串通欺诈者的团伙特别有用？
    • 图像分类
    • 语音识别
    • 共享设备、电子邮件和 IP 的图形分析
    • 拼写检查
  4. What limitation can message-passing GNNs face in imbalanced fraud networks?
    • They guarantee that every linked transaction is fraud
    • They remove the need to monitor changing data
    • They automatically retrain with verified labels
    • Neighbor and hub effects can amplify the dominance of legitimate transactions
  5. Should a fraud model be retrained on a fixed frequent schedule regardless of evidence?
    • No; monitor drift and validate retraining against current data and performance
    • Yes; retraining guarantees fraud patterns will be detected
    • Yes; every model becomes invalid after one day
    • No; changing tactics mean monitoring is unnecessary
  6. What key term does FinCEN ask institutions to include when filing a SAR related to its deepfake alert?
    • FIN-2024-VOICEPRINT
    • FIN-2024-DEEPFAKEFRAUD
    • FIN-2024-ALL-TRANSACTIONS
    • FIN-2024-IDENTITY-GUARANTEE
  7. What does a velocity feature usually measure?
    • The customer’s credit score only
    • The graph’s number of layers
    • The frequency or amount of activity within a specified time window
    • A model’s parameter count
  8. Why should AI fraud alerts be reviewed by investigators?
    • A model score is a court judgment
    • Investigators cannot see transaction data
    • AI prevents all false positives
    • Model outputs may be wrong or incomplete and do not establish legal culpability