Application Of Convolutional Neural Network To Traditional Data

In recent years,convolutional neural networks are widely used in image classification,face recognition,and natural language processing;however,their applications on traditional data remain rare.In this study,traditional data are defined as data whose features have no spatial or temporal dependencies but might have statistical correlations.Such traditional data are also being generated and used in large quantities with the rapid development of information technology.The unique feature extraction feature of convolution neural network makes it have the ability to mine higher dimensional information in data,and the correlation between features of traditional data may be reflected in higher dimensions.Therefore,it is necessary to apply convolutional neural network to traditional data.Aiming at the traditional data classification task,this paper constructs a convolutional neural network model based on feature grid,namely FGCN(Feature Grid-based ConvNet)model.The model consists of two parts:the first part is the feature grid conversion,which converts the traditional data in the form of one-dimensional vector into two-dimensional or higher-dimensional grid data.Two data conversion methods are put forward in this study,which are the optimal grid method and the stochastic grid method.The optimal grid method is inspired by the structural equation model,and the random grid is inspired by the random forest idea.The second part is the model construction,which is based on the transformed data to build a CNN classifier suitable for traditional data.The experimental results show that,in 11 traditional data sets,compared with ten classical classification algorithms,the FGCN model has good performance,so the FGCN model is worth considering for the classification task of traditional data.The predictive application of FGCN model on credit card fraud data shows the value and significance of the model in the field of finance and management.