Detecting Fraudulent Bank Account Based on Convolutional Neural Network with Heterogeneous Data

Figure 1

NTD-CNN classification model inputs with the local topological matrices of bank accounts. There are two convolution layers and each is followed by a pooling layer. denotes the first convolution layer, which contains kinds of convolution kernels. Each kernel, in size of , generates feature maps. represents the first pooling layer, which reshapes the size of a feature map but does not change the number of feature maps. The window size, in , is . In , each kind of input feature map is calculated by kinds of convolution kernels, in which each contains convolution kernel. means the fully connected layer and is the output layer.