Optimizing credit card fraud detection with random forests and SMOTE.

Area of Science:

• Financial Technology
• Machine Learning
• Data Science

Background:

• Credit card fraud is a growing concern in banking, leading to significant financial losses.
• Existing research often focuses on conventional Machine Learning (ML) models, with less emphasis on Deep Learning (DL) techniques for fraud detection.
• The increasing use of credit cards contributes to a rise in default rates, necessitating advanced detection methods.

Purpose of the Study:

• To develop and evaluate a machine learning-based system for detecting fraudulent credit card transactions.
• To compare the efficacy of Deep Learning (DL) models against traditional ML models (Decision Tree, Adaboost) for predicting credit card default.
• To identify key DL parameters that enhance the accuracy of credit card default prediction.

Main Methods:

• Utilized a publicly available, highly imbalanced credit card transaction dataset (fraudulent transactions < 0.2%).
• Employed Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance.
• Evaluated various classification algorithms including Decision Tree (DT), Adaboost, Artificial Neural Networks (ANN), Logistic Regression, and Random Forest, with hyperparameter tuning and Exploratory Data Analysis (EDA).

Main Results:

• The Random Forest model demonstrated superior performance in detecting fraudulent transactions, achieving 99.5% accuracy and a high recall score.
• For credit card default prediction, Adaboost and Decision Tree (DT) models achieved the highest accuracy at 82%, outperforming the Artificial Neural Network (ANN) model (78%).
• The study identified specific DL parameters that contribute to improved accuracy in credit card default prediction.

Conclusions:

• The Random Forest algorithm is highly effective for real-time credit card fraud detection due to its high accuracy and recall.
• While Adaboost and DT show promise for default prediction, further research into DL parameter optimization is needed for enhanced default detection.
• The developed machine learning system can be deployed to bolster fraud prevention mechanisms in financial systems.