Instance-level explanations for fraud detection: a case study

D.A.C. Collaris; Leo M. Vink; J.J. van Wijk

Instance-level explanations for fraud detection: a case study

D.A.C. Collaris, Leo M. Vink, J.J. van Wijk

Research output: Contribution to conference › Paper › Academic

39 Downloads (Pure)

Abstract

Fraud detection is a difficult problem that can benefit from predictive modeling. However, the verification of a prediction is challenging; for a single insurance policy, the model only provides a prediction score. We present a case study where we reflect on different instance-level model explanation techniques to aid a fraud detection team in their work. To this end, we designed two novel dashboards combining various state-of-the-art explanation techniques. These enable the domain expert to analyze and understand predictions, dramatically speeding up the process of filtering potential fraud cases. Finally, we discuss the lessons learned and outline open research issues.

Original language	English
Pages	28-33
Number of pages	6
Publication status	Published - 19 Jun 2018
Event	2018 ICML Workshop on Human Interpretability in Machine Learning (WHI 2018) - Stockholmsmässan, Stockholm, Sweden Duration: 14 Jul 2018 → 14 Oct 2018 Conference number: 3 https://sites.google.com/view/whi2018

Workshop

Workshop	2018 ICML Workshop on Human Interpretability in Machine Learning (WHI 2018)
Abbreviated title	WHI 2018
Country/Territory	Sweden
City	Stockholm
Period	14/07/18 → 14/10/18
Other	Part of W17 of IJCAI-ECAI 2018
Internet address	https://sites.google.com/view/whi2018

Keywords

Interpretability
Explanation
Machine learning
Sensitivity analysis
Local rule extraction
Instance-level explanations
Fraud detection
Case study

Access to Document

arXiv 1806.07129Final author version , 1.49 MB

https://arxiv.org/abs/1806.07129

Cite this

@conference{35e3dbbcdace496c9de10ba20024fe73,

title = "Instance-level explanations for fraud detection: a case study",

abstract = "Fraud detection is a difficult problem that can benefit from predictive modeling. However, the verification of a prediction is challenging; for a single insurance policy, the model only provides a prediction score. We present a case study where we reflect on different instance-level model explanation techniques to aid a fraud detection team in their work. To this end, we designed two novel dashboards combining various state-of-the-art explanation techniques. These enable the domain expert to analyze and understand predictions, dramatically speeding up the process of filtering potential fraud cases. Finally, we discuss the lessons learned and outline open research issues.",

keywords = "Interpretability, Explanation, Machine learning, Sensitivity analysis, Local rule extraction, Instance-level explanations, Fraud detection, Case study",

author = "D.A.C. Collaris and Vink, {Leo M.} and {van Wijk}, J.J.",

year = "2018",

month = jun,

day = "19",

language = "English",

pages = "28--33",

note = "2018 ICML Workshop on Human Interpretability in Machine Learning (WHI 2018), WHI 2018 ; Conference date: 14-07-2018 Through 14-10-2018",

url = "https://sites.google.com/view/whi2018",

}

TY - CONF

T1 - Instance-level explanations for fraud detection: a case study

AU - Collaris, D.A.C.

AU - Vink, Leo M.

AU - van Wijk, J.J.

N1 - Conference code: 3

PY - 2018/6/19

Y1 - 2018/6/19

N2 - Fraud detection is a difficult problem that can benefit from predictive modeling. However, the verification of a prediction is challenging; for a single insurance policy, the model only provides a prediction score. We present a case study where we reflect on different instance-level model explanation techniques to aid a fraud detection team in their work. To this end, we designed two novel dashboards combining various state-of-the-art explanation techniques. These enable the domain expert to analyze and understand predictions, dramatically speeding up the process of filtering potential fraud cases. Finally, we discuss the lessons learned and outline open research issues.

AB - Fraud detection is a difficult problem that can benefit from predictive modeling. However, the verification of a prediction is challenging; for a single insurance policy, the model only provides a prediction score. We present a case study where we reflect on different instance-level model explanation techniques to aid a fraud detection team in their work. To this end, we designed two novel dashboards combining various state-of-the-art explanation techniques. These enable the domain expert to analyze and understand predictions, dramatically speeding up the process of filtering potential fraud cases. Finally, we discuss the lessons learned and outline open research issues.

KW - Interpretability

KW - Explanation

KW - Machine learning

KW - Sensitivity analysis

KW - Local rule extraction

KW - Instance-level explanations

KW - Fraud detection

KW - Case study

M3 - Paper

SP - 28

EP - 33

T2 - 2018 ICML Workshop on Human Interpretability in Machine Learning (WHI 2018)

Y2 - 14 July 2018 through 14 October 2018

ER -

Instance-level explanations for fraud detection: a case study

Abstract

Workshop

Keywords

Access to Document

Fingerprint

Cite this