CREDIT CARD FRAUD DETECTION

BSDSA Project- Fall 2023

Leonardo Dusini, Edoardo Putignano, Stefana Chiriac ,Néstor González

INTRODUCTION TO CREDIT CARD FRAUD DETECTION

• WHAT´S CREDIT CARD FRAUD DETECTION
• It´is a process that uses machine learning techniques to identify suspicious or fraudulent credit card transactions.

​

• WHY IS IT IMPORTANT?
• IMPORTANT FOR THE CLIENTS
• Avoid financial losses and potential disputes for cardholders
• Preserving customer confidence
• IMPORTANT FOR FINANCIAL INSTITUTIONS
• Minimize financial losses
• IMPORTANT FOR THE ECONOMY IN GENERAL

​

DATASET

• UNBALANCED
• VARIABLES
• Time
• time elapsed since the first transaction in the dataset
• V1,V2,...V28
• special features that help the model understand the data without revealing private details.
• ​
• Amount
• amount of the transaction made with the credit card.
• Class
• whether the transaction is legitimate (0) or fraudulent (1).

​

• LOW CORRELATION between variables

Logistic regression

We splitted the data into a train_set and a test_set.

​

We then trained the logistic regression model on the train_set.

Evaluation of the LG model

We measured the accuracy and precision of our first model on the test_set:

​

• Accuracy score: 0.99913

​

• Roc_auc_score: 0.704

New balanced dataset

New logistic regression, trained on a new balanced dataset:

​

• evaluation of the new model on the balanced dataset

accuracy_score : 0.9522

roc_auc_score : 0.924

​

• evaluation of the new model on the balanced dataset

accuracy_score : 0.952

roc_auc_score : 0.926

Random forest

First, we split the data.

Cross validation with K-fold on unbalanced dataset.

​

Random forest

Cross validation with F1-metric:

ROC Curve:

Random forest

Balanced data set (492 and 492)

Cross validation with accuracy metric

Cross validation with F1 metric

Random forest

roc_auc_score:

Conclusion

In the end, comparing the roc_auc_score of our models, we found out that the random forest is the best model to predict future frauds.