class-imbalanced-data-creditcard-transaction

The research was made for the studies of Computer Engineering with a specialization in Computer Graphics and Multimedia Systems at the Wrocław University of Science and Technology. The goal of the research was to create a predictive model for classifying credit card transactions as fraud or a regular transaction. Class-imbalanced data was used to create the model.

Creditcard dataset

Data is CSV file with anonymized credit card transactions labeled as fraudulent or genuine. The datasets contains transactions made by credit cards in September 2013 by european cardholders. This dataset presents transactions that occurred in two days, where we have 492 frauds out of 284,807 transactions. The dataset is highly unbalanced, the positive class (frauds) account for 0.172% of all transactions. It contains only numerical input variables which are the result of a PCA transformation. Due to confidentiality issues, the authors do not provide the original features and more background information about the data. Features V1, V2, ... V28 are the principal components obtained with PCA. The only features which have not been transformed with PCA are 'Time' and 'Amount'. Feature 'Time' contains the seconds elapsed between each transaction and the first transaction in the dataset. The feature 'Amount' is the transaction Amount. Feature 'Class' is the response variable and it takes value 1 in case of fraud and 0 otherwise. The dataset has been collected and analysed during a research collaboration of Worldline and the Machine Learning Group of ULB (University Libre de Bruxelles).

Columns:

Time Number of seconds elapsed between this transaction and the first transaction in the dataset
V1 may be result of a PCA Dimensionality reduction to protect user identities and sensitive features (v1-v28)
V2
V3
V4
V5
V6
V7
V8
V9
V10
V11
V12
V13
V14
V15
V16
V17
V18
V19
V20
V21
V22
V23
V24
V25
V26
V27
V28 abc
Amount Transaction amount
Class 1 for fraudulent transactions, 0 otherwise

Research

1. Different Algorithms

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	99.92%	75.93%	99.96%
Random Forest	99.95%	77.16%	99.99%
MLP	99.75%	78.40%	99.79%
SVM	99.90%	51.85%	99.99%
Logistic regression	99.91%	60.49%	99.98%
KNN	99.83%	4.32%	99,99%
Naive Bayes	99.29%	62.96%	99.35%

2. Feature Selection

2.1. RFECV

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	99.92%	75.93%	99.96%
Random Forest	99.95%	77.16%	99.99%
MLP	99.75%	77.16%	99.79%

2.2. Forward Logistic Regression

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	99.91%	70.99%	99.96%
Random Forest	99.96%	78.40%	99.99%
MLP	99.93%	60.49%	99.99%

2.3. PCA

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	99.91%	69.14%	99.96%
Random Forest	99.95%	77.78%	99.99%
MLP	99.88%	45.06%	99.98%

3. Class-imbalanced problem

3.1 Under Sampling

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	90.79%	89.51%	90.79%
Random Forest	97.96%	89.51%	97.98%
MLP	1.04%	99.99%	0.86%

3.2 Over Sampling

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	99.92%	75.31%	99.97%
Random Forest	99.95%	78.40%	99.99%
MLP	97.25%	90.74%	97.26%

3.3 Smote

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	99.77%	76.54%	99.81%
Random Forest	99.95%	82.10%	99.98%
MLP	99.75%	85.19%	99.78%

3.4 Cluster Centroids

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	26.09%	98.15%	25.97%
Random Forest	6.25%	99.38%	6.09%
MLP	99.70%	20.99%	99.84%

3.5 SMOTETomek

Algorithm	accuracy	sensitivity(TPR)	specificity(TNR)
Decision Tree	99.79%	75.31%	99.84%
Random Forest	99.95%	82.72%	99.98%
MLP	99.33%	86.42%	99.35%

adamek195/class-imbalanced-data-creditcard-transaction