Big Data Analytics: Fraud Detection

📝 Description

The project presents a proof of concept of a Fraud Detection Analytics Solution. The purpose of the project to demonstrate how the potential fraud can be detected based on the mockup data set reflecting the medical records and generated with mockaroo.com and also include the public datasets from kaggle.com.

---

Tools / Languages

Platform

Data Sources

• mockaroo.com - lets you generate up to 1,000 rows of realistic test data in CSV, JSON, SQL, and Excel formats
• kaggle.com - Access a huge repository of community published data & code

---

1️⃣ Data Model

2️⃣ Number of Double-Payments per Provider vs Number of Total Claims per Provider

y = 0.015x + 12.757
R² = 0.8291 r=0.910549285 Strong Positive Correlation

> DPvsProv<-read.csv("DPvsProv.csv", header=TRUE)

> DPvsProv
PROVIDER_ID PROV_CLM_COUNT D_FLAG

1  0  1421  146
2  3  46  0
3  11  36256  964
4  17  10119  223

plot(DPvsProv$PROV_CLM_COUNT, DPvsProv$D_FLAG, xlab = "Total number of claims per provider", ylab = "number of double-payments per provider", pch = 16, cex = 1.3, col = "blue")

with(DPvsProv,plot(DPvsProv$PROV_CLM_COUNT, DPvsProv$D_FLAG, xlab = "Total number of claims per provider", ylab = "number of double-payments per provider", pch = 16, cex = 1.3, col = "blue"))

3️⃣ Number of Wrong ICDs per Provider vs Total Number of Claims per Provider

y = 0.0085x - 51.938
R² = 0.23 r=0.479583152 Weak Positive Correlation

ICDsvsProv<-read.csv("ICDsvsProv.csv", header=TRUE)

ICDsvsProv
PROVIDER_ID PROV_CLM_COUNT WRONG_ICD

1  0  1421  2
2  3  46  0
3  11  36256  145
4  17  10119  9
5  25  26712  42
6  45  843  0
7  56  128418  5173

plot(ICDsvsProv$PROV_CLM_COUNT, ICDsvsProv$WRONG_ICD, xlab = "Total number of claims per provider", ylab = "Number of Wrong ICDs per Provider", pch = 16, cex = 1.3, col = "blue")

with(ICDsvsProv,plot(ICDsvsProv$PROV_CLM_COUNT, ICDsvsProv$WRONG_ICD, xlab = "Total number of claims per provider", ylab = "Number of Wrong ICDs per Provider", pch = 16, cex = 1.3, col = "blue"))

4️⃣ Total Number of Claims vs Double-Payments per Location

y = 0.0172x - 47.984
R² = 0.9147 r=0.956399498 Strong Positive Correlation

DPvsLoc<-read.csv("DPvsLoc.csv", header=TRUE)

DPvsLoc
PROV_LOCATION_ID LOC_CLM_COUNT D_FLAG

1  1  329879  4809
2  2  268034  3779
3  3  283200  6040
4  4  107953  1925
5  5  94588  1607
6  6  582  18
7  7  173951  1200

plot(DPvsLoc$LOC_CLM_COUNT, DPvsLoc$D_FLAG, xlab = "Total number of claims per location", ylab = "number of double-payments per location", pch = 16, cex = 1.3, col = "blue")

with(DPvsLoc,plot(DPvsLoc$LOC_CLM_COUNT, DPvsLoc$D_FLAG, xlab = "Total number of claims per location", ylab = "number of double-payments per location", pch = 16, cex = 1.3, col = "blue"))

5️⃣ Total Number of Claims per Location vs Number of Wrong ICDs per Location

y = 0.0043x - 34.453
R² = 0.8262 r=0.908955444 Strong Positive Correlation

ICDsVSLoc<-read.csv("ICDsVSLoc.csv", header=TRUE)

ICDsVSLoc
location_ID LOC_CLM_COUNT WRONG_ICD
PROV_LOCATION_ID LOC_CLM_COUNT WRONG_ICD

1  1  329879  1205
2  2  268034  1208
3  3  283200  1023
4  4  107953  814

plot(ICDsVSLoc$LOC_CLM_COUNT, ICDsVSLoc$WRONG_ICD, xlab = "Total number of claims per location", ylab = "Number of Wrong ICDs per location", pch = 16, cex = 1.3, col = "blue")

with(ICDsVSLoc,plot(ICDsVSLoc$LOC_CLM_COUNT, ICDsVSLoc$WRONG_ICD, xlab = "Total number of claims per location", ylab = "Number of Wrong ICDs per location", pch = 16, cex = 1.3, col = "blue"))