Fighting against Organized Fraudsters Using Risk Diffusion-based Parallel Graph Neural Network

1. Code

Coming soon.

2. Medical Fraud Dataset

2.1 PURPOSE

We found that the data about medical fraud is rare and it is difficult to search accurate and reliable information for it, so we collected some medical fraud data and give fraud labels and gang fraud labels according to experts in relevant fields to help follow-up researchers to study this field and relieve the lack of data.

2.2 FORMAT DESCRIPTION

In the dataset, the format of the date is : year/month/day hour:minute:second. We convert the null values in original data to -1.

2.3 STATISTICAL INFORMATION

• about 550 thousands claims
• data from 2009
• about 5000 providers,500 fraud providers,100 group fraud providers
• 138 thousands beneficiaries

2.4 DATA SOURCE

The origin source of the data comes from the website and you can find more information about it on :https://www.kaggle.com/datasets/rohitrox/healthcare-provider-fraud-detection-analysis

2.5 THE FORMAT AND THE MEANINGS

claim.csv

attribute	meaning
ClaimID
BeneID
ClaimStartDt	Date of claim start
ClaimEndDt	Date of claim end
InscClaimAmtReimbursed	Claim amount
AdmissionDt	Date of admission
ClmAdmitDiagnosisCode	Claim diagnostic code
DeductibleAmtPaid	Patient self payment
DischargeDt	Date the patient was discharged
DiagnosisGroupCode	Diagnostic classification group code for patients
inpatent	Whether this claim is inpatient
PotentialFraud	Fraud label
PotentialGroupFraud	Group Fraud label

bene.csv

attribute	meaning
BeneID
DOB	Date of birth
DOD	Date of death
Gender
Race
RenalDiseaseIndicator	the patient have renal failure problems
State
County
NoOfMonths_PartACov
NoOfMonths_PartBCov
ChronicCond_Alzheimer	Have Alzheimer disease (2 no, 1 yes)
ChronicCond_Heartfailure	Have Heartfailure (2 no, 1 yes)
ChronicCond_KidneyDisease	Have KidneyDisease (2 no, 1 yes)
ChronicCond_Cancer	Have Cancer (2 no, 1 yes)
ChronicCond_ObstrPulmonary	Have ObstrPulmonary (2 no, 1 yes)
ChronicCond_Depression	Have Depression (2 no, 1 yes)
ChronicCond_Diabetes	Have Diabetes (2 no, 1 yes)
ChronicCond_IschemicHeart	Have IschemicHeart (2 no, 1 yes)
ChronicCond_Osteoporasis	Have Osteoporasis (2 no, 1 yes)
ChronicCond_rheumatoidarthritis	Have rheumatoidarthritis (2 no, 1 yes)
ChronicCond_stroke	Have stroke (2 no, 1 yes)
IPAnnualReimbursementAmt	Maximum reimbursement amount for hospitalization
IPAnnualDeductibleAmt	patient's annual hospital deductible
OPAnnualReimbursementAmt	Maximum reimbursement amount for outpatient visits
OPAnnualDeductibleAmt	The patient's annual out of pocket expenses for outpatient visits

3. Citing

If you find RDPGL is useful for your research, please consider citing the following papers:

@inproceedings{mafighting,
    title={Fighting against Organized Fraudsters Using Risk Diffusion-based Parallel Graph Neural Network},
    author={Ma, Jiacheng and Li, Fan and Zhang, Rui and Xu, Zhikang and Cheng, Dawei and Ouyang, Yi and Zhao, Ruihui and Zheng, Jianguang and Zheng, Yefeng and Jiang, Changjun},
    booktitle={International Joint Conference on Artificial Intelligence},
    year = {2023},
    month = {08},
    pages = {6138-6146},
    doi = {10.24963/ijcai.2023/681}
}