Detection of anomalous transactions in the blockchain with adaptive multi-feature fusion

Abstract

With the goal of addressing the issue that the original data's (features') representation ability was limiting the performance of intelligent detection models, a residual network structure called ResNet-32 was created. ResNet-32's purpose is to actively learn the high-level abstract features with rich semantic information by automatically mining the complex association relationships between original features.While low-level features were less able to differentiate themselves from high-level features, they were nevertheless more descriptive of transaction content.Enhancing the detection performance required knowing how to combine them to acquire complimentary advantages.