Kolaboro / Risk patterns
Risk patterns: unusual versus peers
Which tracked projects stand out on public warning signs, and exactly which signs. Every point of every score traces to one feature. It ranks projects against each other; it does not predict fraud, because no outcome data exists to test such a claim.
…unusual (top 5% of scores)
…somewhat unusual (top 25%)
…typical
…rank agreement with an isolation forest (Spearman)
Explain one project
The most unusual projects
The twelve highest scores. Ids link to each project's public page. A project here has rare warning signs compared with the others we track; it has not been found guilty of anything.
Model card
- Version
- …
- Built
- …
- Projects
- …
- Dataset date
- …
- Unusual from score
- …
- Somewhat unusual from
- …
- Top 5% in both methods
- …
Data: /data/bali-cases.json, SHA-256 …
Purpose:
Not for:
Features
| Feature | Kind | Share of projects | Robust z-score |
|---|
Wanted but not in the model yet
Method
Limits
Median score by market (markets with 8 or more projects)
| Market | Projects | Unusual | Median score |
|---|
Data: /data/risk.json · API: /api/x/h/risk, /api/x/h/project/{id} · Related: advertised versus registered prices, automated register checks.