Intelligent Algorithms for the Detection of Suspicious Transactions in Payment Data Management Systems Based on LSTM Neural Networks

TL;DR

This paper explores using LSTM neural networks and an optimization algorithm to detect suspicious transactions in payment systems, aiming to improve accuracy and efficiency in fraud detection.

Contribution

The novel integration of an Artificial Bee Colony optimization algorithm with LSTM networks for hyperparameter tuning in detecting suspicious transactions.

Findings

LSTM-based models with ABC optimization outperformed traditional models in detecting suspicious transactions.

The ABC algorithm improved the accuracy and efficiency of the LSTM model in identifying complex fraud patterns.

Preliminary data preparation significantly enhanced the performance of intelligent detection methods.

Abstract

Today, a number of works are being carried out all over the world to develop data processing and management systems, as well as to apply artificial intelligence and information technologies in the fields of production, science, education, and healthcare. The optimization of the management of socio-economic process systems, and the management and reliability of databases of the digital payment information-based information systems of enterprises and organizations are relevant. This research work investigates the issue of increasing the reliability of information in information systems working with payment information. The characteristics of ambiguous suspicious transactions in payment systems are analyzed, and based on the analysis, preliminary data preparation stages are carried out for the intelligent detection of ambiguous suspicious transactions. Traditional and neural network models of machine learning for the detection of suspicious transactions in payment information management systems are developed, and a comparative analysis is carried out. Furthermore, to enhance the performance of the core LSTM model, an Artificial Bee Colony (ABC) optimization algorithm was integrated for automated hyperparameter tuning, which improved the model’s accuracy and efficiency in identifying complex fraudulent patterns.

Figures