引導式測驗 · 中等 等級

AI在公共預算預測測驗

測試人工智慧預算預測:目標和範圍、時間感知驗證、透明假設、不確定性範圍、基準和公共責任。

相關引導路徑AI在公共預算預測
問題 1 的 8

A revenue model is trained on five years of records. Which test best approximates forecasting next year?

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. A revenue model is trained on five years of records. Which test best approximates forecasting next year?
    • Randomly split rows from all six years
    • Evaluate only on the training records
    • Train on earlier periods and evaluate on a later period kept out of model fitting
    • Choose the year with the lowest error after viewing all results
  2. A model predicts annual tax receipts. Why must the office state its target and horizon?
    • The target and time horizon define what the estimate means and how it can guide decisions
    • They make an uncertain forecast certain
    • They determine which political priorities should be funded
    • They eliminate data revisions
  3. A tax-law change breaks a historical relationship in the data. What is the main risk?
    • The model automatically learns the new law before it takes effect
    • The change improves accuracy by adding more data
    • The forecast horizon becomes irrelevant
    • The model may extrapolate a pattern that no longer describes the revenue system
  4. Why compare a complex forecast with a simple baseline?
    • To make the complex model appear more accurate
    • To determine whether added complexity improves useful performance over a transparent reference
    • Because a baseline guarantees the correct budget
    • To avoid testing on future periods
  5. Which presentation best communicates uncertainty to budget decision-makers?
    • Publish only the model’s most likely number
    • Hide sensitivity analysis to prevent debate
    • Show a central estimate with plausible ranges or scenarios and explain key assumptions
    • Convert the estimate into a guarantee
  6. A random train-test split gives excellent results on monthly revenue data. What concern remains?
    • The result proves the budget will be balanced
    • Random splitting ensures a realistic future simulation
    • The model has no need for a baseline
    • Future observations may have leaked into training, making the test unlike a true forecast
  7. A city sees its average error fall but the model repeatedly misses during downturns. What should analysts report?
    • Performance by relevant conditions, including downturns and large misses
    • Only the overall average error
    • 模型的參數數量
    • That no forecast can be evaluated
  8. Who remains responsible for selecting spending priorities after an AI forecast is delivered?
    • The forecasting model because it produced the estimate
    • Public officials following the applicable budget process and considering public priorities
    • The vendor that selected the algorithm
    • The department with the largest dataset