Research on data anomaly detection algorithm for financial audit
To better promote the digitalization of audit and fulfil the task of detecting data anomalies in financial audits,a CMA-Resnet18 model based on an improved attention mechanism CMA(Channel Mixed Attention mechanism)of independent research is designed,and a detection dataset construction method based on the idea of digital image transformation is proposed.First,the CMA network is employed to globally weight each channel of the sample and weight different channels of the sample with fusion features to realize the global"attention"data enhancement of the sample data.Then,the local features of the sample data are extracted by Resnet18 model(residual network18).Finally,the intelligent detection of financial data anomalies is realized.Our experimental results show on the financial audit anomaly detection dataset,the evaluation of the classical classification network is higher than 90%,proving the effectiveness of the dataset construction method;the F1 value of the CMA-Resnet18 model is 94.31%,1.49%higher than that of Resnet18,demonstrating the model performs better in the detection tasks.Our experiments with classical classification networks and their CMA variants on the Cifar10 public dataset show the accuracy of the CMA variants is generally higher than that of their original networks,demonstrating the effectiveness and generalization of the CMA module.
国家科技期刊平台
NSTL
万方数据
