Project Name

Credit Card Fraud Detection (Group Project)

Summary

Given a dataset of credit card transactions, we performed preprocessing, resampling, and supervised learning to predict fraudulant transactions. We also compared models based on accuracy, precision, recall and f1-score.

Details can be found in documents/Final Report.pdf.

Dataset

IEEE-CIS Fraud Detection: https://www.kaggle.com/c/ieee-fraud-detection/discussion/101203

1. Preprocessing

Dropping data

• Remove features not apprpriate for in our process such as Transaction ID and TransactionDT (delta time)
  
• Remove columns with more than a threashold-percent of NaNs
• Fill NaN in columns with column means
  

Dimensionality reduction

• After min-max normalization, remove features with variance lower than threshold since features essentially constant do not help the fitting.
  
• Remove features with high correlation to other features since it can be redundant information when fitting
  
• Perform PCA and standardize data beforehand (if a feature shows high variance, this may dominate the principal components).
  

PCA

• Standardize data (mean 0 variance 1)
  
• Perform PCA with scikit-learn
  

2. Resampling with SMOTE

The givien data is skewed towards transactions which are not fradulent, we will use SMOTE to sinthesize more fradulent transation data points to have a 1:1 ratio of the classes.

3. Supervised learning

Applied kNN, NBayes, Linear Regression, SVM, Neural Networks (FFN, CNN), Random Forest, and XG Boost to predict fraudulant transactions.