Dennis Collaris
Computer Science & Engineering
Uitlegbaarheid van
Machine Learning modellen
Afstudeer project
Uitlegbaarheid van
Machine Learning modellen
Machine Learning
Fraud detectie
bij ziekte verzuim verzekeringen
Fraud detectie model
75% Fraude
Fraude
Detectie
Model
Fraude team
| Company | ABC Inc |
| Employees | 5 |
| Illness duration | 14 days |
| Premium rate | 5% |
| ... | ... |
Verzekeringspolis
Uitleg
Aha!
Duration illness < 14 days
Non-fraud
Yes
No
Premium percentage < 5%
Yes
No
Non-fraud
Fraud
Modellen
-
Decision Tree
-
Random Forest
- 100 Random Forests
(ensemble)
Beslissingen:
2
23
69
12,704
1,312,471
Lastig probleem
Globaal vs Locaal
Algemeen:
Datum melding is verdacht
Voor deze werkgever: Duur Ziekte is belangrijk
Globaal vs Locaal
Mijn oplossing
Dashboards
Feature importance
Technique 1
Techniek 2
Techniek 3
1. Feature importance
| Company | ABC Inc |
| Employees | 5 |
| Illness duration | 14 days |
| Premium rate | 5% |
| ... | ... |
Verzekeringspolis
Techniek 2
Techniek 3
1. Feature importance
Techniek 2
Techniek 3
1. Feature importance
"Onenigheid"
Techniek 2
Techniek 3
1. Feature importance
Demo
Techniek 2
Techniek 3
1. Feature importance
Sensitivity analysis
Technique 2
2. Sensitivity analysis
Technique 3
1. Feature importance
| Company | ABC Inc |
| Employees | 5 |
| Illness duration | 14 days |
| Premium rate | 5% |
| ... | ... |
Verzekeringspolis
300
250
200
150
100
50
1
0%
100%
300
200
100
0
DuurZiekte
Fraude?
Fraude (55%)
Geen fraude (35%)
| Company | ABC Inc |
| Employees | 5 |
| Illness duration | days |
| Premium rate | 5% |
| ... | ... |
Fraude (65%)
Fraude (90%)
Geen fraude (45%)
Geen fraude (40%)
Geen fraude (25%)
2. Sensitivity analysis
Techniek 3
1. Feature importance
Demo
2. Sensitivity analysis
Techniek 3
1. Feature importance
Model simplificatie
Technique 3
2. Sensitivity analysis
3. Model simplification
1. Feature importance
Complex
Model
| Company | ABC Inc |
| Employees | 5 |
| Illness duration | 14 days |
| Premium rate | 5% |
| ... | ... |
Verzekeringspolis
Simpel
Model
2. Sensitivity analysis
3. Model simplification
1. Feature importance
Policy 1 Fraud (88%)
Policy 2 Non-fraud (25%)
2. Sensitivity analysis
3. Model simplification
1. Feature importance
Demo
2. Sensitivity analysis
3. Model simplification
1. Feature importance