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。
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。
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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