Smart credit card fraud detection system based on dilated convolutional neural network with sampling technique

Karthika, J; Senthilselvi, A.

doi:10.1007/s11042-023-15730-1

Cited by 9 publications

(1 citation statement)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Among such methods deep learning models nowadays provide remarks results in various applications, its notable areas are to analyze medical images, recognize objects, process natural language processing, analyze the structure of genomes in Bioinformatics, and detection of fraudulent transactions. For this purpose many learning methods have been employed by researchers, they are Deep Learning (DL), Reinforcement Learning (RL), and Deep Reinforcement Learning (DRL) [6,7]. The main objective of this research is to detect credit card fraudulent transactions in real-time using deep reinforcement learning.…”

Section: Introductionmentioning

confidence: 99%

A Novel Approach for Credit Card Fraud Transaction Detection Using Deep Reinforcement Learning Scheme

Qayoom

Khuhro

Kumar

et al. 2023

Preprint

View full text Add to dashboard Cite

Online transactions are the backbone of the financial industry globally. Millions of people are using credit cards daily, which has led to an exponential rise in their daily life that has led to an exponential rise in credit card frauds. With the passage of time, many variations and schemes of fraudulent transactions have been occurring. Real-time credit card fraud detection is still a challenging task. Every person has a unique transaction pattern. This pattern may vary over a period of time. Sometimes, credit card frauds have no constant pattern. This paper aims to 1) understand how deep reinforcement learning can play a vital role in detecting credit card fraud and 2) design solution architecture for real-time detection. Our proposed model uses the DeepQ network. A deep Q network has been utilized for real time detection. The online available Kaggle dataset hasbeen used for model training and testing, and 97% validation performance has been achieved via our proposeddeep learning component. The reinforcement learningcomponent has an 80% learning rate. Our proposedmodel was able to learn patterns itself based on previous history and adopted the pattern changes over timeand accommodate them without any further manualtraining.

show abstract

Section: Introductionmentioning

confidence: 99%