Tailoring Data Storage Configuration for Efficient Fraud Detection Model Training

Syarif, Abdusy and Putra, Muhammad Haikal and Gabteni, Hanene (2025) Tailoring Data Storage Configuration for Efficient Fraud Detection Model Training. Jurnal Ilmiah Teknik Elektro Komputer dan Informatika (JITEKI), 10 (4). pp. 1021-1036.

Text 26-Tailoring Data Storage Configuration for Efficient Fraud.pdf Download (1MB)

Abstract

The rapid growth of e-commerce in Indonesia, with a record 88.1% growth rate, has been accompanied by a surge in online fraud, leading to an estimated loss of 4.62 trillion rupiahs. Current fraud prevention methods, such as the widely used 3D-Secure system, though effective, result in a high rate of transaction abandonment (approximately 16%), which is undesirable for merchants. To address this, we propose an AI-based fraud detection system that leverages machine learning models to identify potentially fraudulent transactions. By employing a combination of classification algorithms, including logistic regression and neural networks, security protocols are activated only for high-risk transactions, optimizing transaction processing efficiency and improving detection accuracy. Our study focuses on fine-tuning key parameters of the AI-Fraud Detector model, specifically some parameters such as ∆ttrain, ∆tlag and f rac hr pass, to enhance detection performance over time. Simulation performances using ROCAUC, false positive rate (fpr), and true positive rate (tpr) metrics show that a configuration with a training period (∆ttrain) of 180 days, a lag period (∆tlag ) of 90 days, and a high-risk pass fraction (f rac hr pass) of 10% yields a balance between detection efficiency (∼ 50%) and a reduced false positive rate. It means that the model is able to identify approximately 50% of the actual high-risk events while minimizing the number of times it incorrectly identifies a low-risk event as high-risk. However, further research is required to refine these results, explore parameter optimization strategies, and enhance the model’s adaptability to evolving fraud patterns. Future work will focus on optimizing thresholds, improving model robustness over time, and ensuring effective detection of new fraud schemes. This research improves model performance by optimizing key parameters and enhancing detection accuracy while minimizing false positives

Item Type:	Artikel Umum
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisi / Prodi:	Faculty of Industrial Technology (Fakultas Teknologi Industri) > S1-Electrical Engineering (S1-Teknik Elektro)
Depositing User:	M.Eng. Alfian Ma'arif
Date Deposited:	21 Feb 2025 07:24
Last Modified:	21 Feb 2025 07:24
URI:	http://eprints.uad.ac.id/id/eprint/82035

Actions (login required)

View Item