Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ
Đố vui rừng biệt lập
Kết nối độ dài đường dẫn cây ngẫu nhiên với thứ hạng bất thường và phân biệt ngưỡng ô nhiễm với tỷ lệ bất thường trong thế giới thực.
Câu hỏi 1 của 8
What tends to happen to an unusual point's average isolation-tree path length?
Xem trước mọi câu hỏi8 câu hỏi
Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.
- What tends to happen to an unusual point's average isolation-tree path length?
- It is shorter because fewer random splits isolate it
- It is longer because every split keeps it with common points
- It equals the number of features exactly
- It is set by contamination directly
- What does the contamination setting commonly control?
- The verified prevalence of anomalies in production
- The threshold used to label a chosen fraction as outliers
- The number of input features
- The random split direction for every tree
- Which score interpretation matches the forest's core intuition?
- Longer paths always mean higher anomaly risk
- A score is a calibrated fraud probability
- Shorter paths indicate stronger isolation-based anomaly evidence
- The score identifies the causal reason for an event
- A rare but legitimate customer segment receives high anomaly scores. What issue does this illustrate?
- Path lengths are always causal
- Contamination guarantees correct labels
- Random trees require class labels
- Rarity is not equivalent to fraud or invalidity
- Why verify score direction in the chosen library?
- All libraries define negative scores as normal
- Some decision functions invert or shift the raw anomaly score
- The tree ensemble omits path lengths
- Contamination reverses the feature matrix
- If contamination increases from 5% to 10%, what change is most directly expected?
- Every path becomes shorter
- The true anomaly rate is proven to double
- More points may be labeled anomalous by the threshold
- The model changes into supervised fraud classification
- How should an alerting team evaluate an Isolation Forest threshold when labels are available?
- Choose the threshold that labels exactly half the data
- Assume every short path is fraud
- Compare only the training score average
- Measure precision and recall tradeoffs on representative labeled cases
- What can an Isolation Forest score not provide by itself?
- A causal explanation or calibrated event probability
- A ranking of isolation behavior
- An ensemble-based anomaly score
- A comparison across observations