Class balancing framework for credit card fraud detection based on clustering and similarity-based selection (SBS)

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Credit card fraud is a significant financial crime, exacerbated by the shift to cashless transactions during the COVID-19 pandemic.
• Traditional fraud detection models struggle with imbalanced datasets, leading to inaccurate predictions by favoring the majority class.
• Billions of Euros are lost annually, necessitating advanced fraud detection systems for financial institutions.

Purpose of the Study:

• To improve the accuracy of credit card fraud detection systems.
• To address the challenge of imbalanced datasets in machine learning models for fraud detection.
• To propose an effective under-sampling technique for better fraud prediction.

Main Methods:

• Implemented an under-sampling technique to balance the credit card transaction dataset.
• Utilized fuzzy C-means clustering to group similar instances (fraudulent and normal) based on shared features.
• Developed a framework to ensure data integrity by selecting comparable instances.

Main Results:

• The under-sampling approach, combined with fuzzy C-means clustering, led to more accurate fraud detection.
• The proposed method effectively handled imbalanced data, improving model performance.
• Ensured feature integrity between selected normal and fraudulent data points.

Conclusions:

• Under-sampling techniques, particularly when combined with fuzzy C-means clustering, offer a robust solution for imbalanced credit card fraud detection.
• The developed framework enhances the reliability and accuracy of machine learning models in identifying fraudulent transactions.
• This research provides a valuable method for financial institutions to mitigate losses from credit card fraud.