ADS-503 Applied Predictive Analytics Project

Financial Datasets For Fraud Detection

Leonard Littleton
Lina Nguyen
Emanuel Lucban

---

Downloads

Dataset for the project can be downloaded here
Kaggle - Synthetic Financial Datasets for Fraud Detection

RDS files for all trained models can be downloaded here
Note: zip file is 30GB
rds_files.zip

---

Models Tested

• Quadratic Discriminant Analysis
• Support Vector Machines
• Logistic Regression
• Gradient Boosting Machine
• Linear Discriminant Analysis
• Nearest Shrunken Centroids
• Neural Networks
• Partial Least Squares Discriminant Analysis

---

Model Performance

ROC Curves

Metrics

Models	AUC	Sensitivity	Specificity	Pos.Pred.Value	Neg.Pred.Value	Precision	Recall	F1	Prevalence	Detection.Rate	Detection.Prevalence	Balanced.Accuracy
Quadratic Discriminant Analysis	0.9927164	0.8594569	0.9937371	0.1534927	0.9998132	0.1534927	0.8594569	0.2604678	0.0013196	0.0011341	0.0073888	0.9265970
Support Vector Machines	0.9978020	0.7617913	0.9999610	0.9626731	0.9996853	0.9626731	0.7617913	0.8505319	0.0013196	0.0010052	0.0010442	0.8808762
Logistic Regression	0.9964422	0.7308242	0.9999553	0.9557632	0.9996444	0.9557632	0.7308242	0.8282937	0.0013196	0.0009644	0.0010090	0.8653898
Gradient Boosting Machine	0.9777450	0.6279181	0.9999465	0.9394155	0.9995086	0.9394155	0.6279181	0.7527127	0.0013196	0.0008286	0.0008820	0.8139323
Linear Discriminant Analysis	0.9621525	0.4030491	0.9999037	0.8468468	0.9992118	0.8468468	0.4030491	0.5461588	0.0013196	0.0005319	0.0006280	0.7014764
Nearest Shrunken Centroids	0.9362662	0.0724154	0.9987977	0.0737148	0.9987744	0.0737148	0.0724154	0.0730594	0.0013196	0.0000956	0.0012963	0.5356065
Neural Network	0.5000000	0.0000000	1.0000000	NaN	0.9986804	NA	0.0000000	NA	0.0013196	0.0000000	0.0000000	0.5000000
PLS Discriminant Analysis	0.9771492	0.0000000	1.0000000	NaN	0.9986804	NA	0.0000000	NA	0.0013196	0.0000000	0.0000000	0.5000000

---

References

Lopez-Rojas, Edgar. (2017). Synthetic Financial Datasets For Fraud Detection. Kaggle. https://www.kaggle.com/ealaxi/paysim1.