引導式測驗 · 硬 等級

使用機器學習偵測會計舞弊測驗

了解統計螢幕和機器學習如何標記會計模式、為什麼警報不能作為證據,以及審計員如何評估檢測系統。

問題 1 的 8

What does a high fraud-risk model score establish?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What does a high fraud-risk model score establish?
    • A score flags a pattern for investigation, not proof
    • A pattern is suspicious only if a model labels it fraud
    • A high score confirms that a specific journal entry was intentionally falsified
    • The model score is sufficient evidence for an accusation
  2. When is Benford’s law a poor fit for a dataset?
    • Benford’s test works equally for assigned invoice IDs
    • A restricted amount range makes naturally occurring digit assumptions stronger
    • Assigned identifiers and constrained prices may not fit Benford assumptions
    • A digit mismatch proves that someone fabricated transactions
  3. Why can accuracy be misleading for rare fraud detection?
    • A model with high accuracy must have found enough rare fraud cases
    • Majority-class accuracy can hide missed rare cases
    • Class imbalance guarantees high recall
    • Majority predictions necessarily identify rare manipulation
  4. What does a Beneish M-score provide?
    • A deterministic cutoff with the same interpretation for every industry and period
    • An M-score proves intentional falsification
    • The M-score replaces supporting audit evidence
    • The M-score is a screening signal, not a fraud finding
  5. Why use time-aware evaluation splits for fraud models?
    • Time splits reduce leakage from future information
    • Randomly mix future and past cases to simulate deployment
    • Use later filing revisions as features because they improve historical fit
    • Train on later labels to predict earlier decisions
  6. Which metric estimates the share of inspected alerts that are relevant within a fixed review capacity?
    • Recall across all cases without considering the number staff can inspect
    • Overall accuracy alone predicts review workload
    • Precision at review capacity informs alert workload
    • AUC tells exactly how many staff hours are needed
  7. What should an investigator do with a flagged journal entry?
    • A flagged entry should be removed before audit review
    • Trace an entry to documents and assess business context
    • Compare only with peer companies and skip the underlying source records
    • An unusual entry is proof of intent
  8. Why might a legitimate transaction appear anomalous?
    • A historical average match makes a transaction valid regardless of context
    • A baseline deviation always indicates an error in the training data
    • Only misconduct produces deviations from a baseline
    • Legitimate activity can be unusual relative to a baseline