The Parallel GRU -LSTM Neural Networks for Accurate Detection of Fraudulent Bitcoin Transactions

The Financial fraud continues to rise alarmingly, even as technological advancements strive to address it. This paradoxical trend stems primarily from inadequate coordination among organizations and privacy concerns, which impede access to reliable and comprehensive transaction data. To tackle this challenge, this study introduces an innovative blockchain-based framework designed to detect fraud effectively by leveraging cutting-edge technologies. The proposed system integrates deep learning and machine learning techniques into a cohesive model, enabling robust analysis of transactional data. By constructing a detailed graph representation of sequential transactions on the Bitcoin blockchain, the model identifies patterns, extracts meaningful features, and classifies them to detect suspicious or fraudulent activities. Central to this approach is the Parallel Model, which combines two advanced sequential architectures, namely the Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM). These architectures work collaboratively to capture complementary features from time-series input data, ensuring a comprehensive understanding of the temporal and sequential patterns inherent in blockchain transactions. Additionally, the system incorporates the XGBoost algorithm, a powerful feature selection technique, to enhance classification performance. This integrated approach optimizes the model’s accuracy in identifying fraudulent transactions, offering a scalable and reliable solution to combat financial fraud in blockchain ecosystems.

Keywords:

Bitcoin , Blockchain , Fraud detection , Deep learning , Machine learning , Deep recursive neural network , Parallel model