A Threat Detection Scheme for Financial Big Data in Internet of Things

Junzhe Jia^1*Li Zhou²

• ¹Xidian University, Xi'an, China
• ²McGill University, Montreal, Canada

With the deep application of Internet of Things (IoT) technology in the financial field, the transmission, storage and processing of massive financial data face complex and diverse security threats. This paper proposes a threat detection scheme, CNN - BiLSTM - GAM, which is based on the vulnerabilities of IoT devices in financial big data scenarios and deep learning algorithms. By analyzing the traffic data and behavioral patterns generated by IoT devices during data collection and other processes, it extracts key features and identifies security threats such as malicious attacks. CNN-BiLSTM-GAM includes Convolutional Neural Network (CNN), Bidirectional long short-term memory (BiLSTM) and global attention module (GAM), which accurately extract spatial features of input financial data through one-dimensional convolutional neural network (1D-CNN). At the same time, BiLSTM layer captures the context dependency relationship in time series data through forward and backward networks. It optimizes the extraction of temporal features, finally assigns weights to input features through the global attention obtained by concatenating channel attention and spatial attention. The experimental results show that CNN-BiLSTM-GAM performs well with 96.81% of ACC and 96.79% of F1 on NSL-KDD, 96.98% of ACC and 96.46% of F1 on CICIDS2017, demonstrating better spatiotemporal feature extraction capabilities and providing technical support for ensuring the security of financial big data.