Positive-Unlabeled (PU) learning — when useful?
hardAnswer
- You have some confirmed positives (frauds, anomalies) and a large unlabeled pool that mixes positives + negatives.
- Standard classification is biased.
- PU methods: (1) treat unlabeled as noisy negatives and correct via propensity, (2) two-step: identify reliable negatives then train classifier, (3) unbiased PU risk estimator (Kiryo et al. 2017).
- Uses: fraud (few labeled frauds, most 'clean' actually clean), rare-disease diagnosis.
Check yourself — multiple choice
- Random
- Learn from confirmed positives + unlabeled pool (mix of pos + neg); use propensity correction / two-step / unbiased PU risk (Kiryo); fraud, rare-disease
- Same as supervised
- Not real
PU learning: positives + unlabeled; propensity / two-step / unbiased risk.
#anomaly-detection
Practise Unsupervised Learning
214 interview questions in this topic.
Related questions
- What are the main approaches to anomaly detection?
- How do outliers affect k-means and how do you handle them?
- Isolation Forest — how does it detect anomalies?
- Local Outlier Factor (LOF) — how does it work?
- One-class SVM — mechanism and pitfalls.
- Elliptic envelope / robust Mahalanobis distance — when to use?