Research On Risk Control Of Networked Audit Based On Data Mining Technology

In an era of global competition, market changes and information explosion, the auditor’s responsibility is to obtain useful data from the large amount of data contained in the data warehouse of the audit unit, and through the analysis of data to publish appropriate audit opinion. On the one hand, the audit object from the previous paper-based books into electronic data types and formats are complex and also makes audit personnel to undertake the audit responsibility and facing the audit risk is also increasing. On the other hand, the enterprise management is more concerned about the innovation of management strategy, and the related risk management and control. At the same time, the function and position of audit is also quietly changing, from the traditional focus to the attention and report certain threat factors or risk factors, from the inspection of financial statements to master the entire operating system and risk factors, to find a powerful and relatively low risk of audit methods, is the development trend of modern audit research.Through consulting a large number of newspapers, journals, magazines, and reading a lot of literatures from data mining and online audit, the paper further explores the audit risk and audit quality control. Finally, the three chapter constructs a commercial bank’s risk control system, and the bank’s credit card marketing data is applied to build the internal audit data analysis, data mining(decision tree, gradient Boosting, regression, neural network, automatic neural network) model to deal with data. We set up the 60% part of the audit data set into a training set, the 40% part is set to become a validation set, and assume that the former card of these people, if successful marketing can be profitable 95 yuan, if the failure is 5 yuan, and finally dig out the ideal results. Through the observation of five kinds of data mining model of ROC, the use of decision tree data processing effect is best, and the five models in the processing of data above both training set and validation set of the above results are similar, we take marketing success probability is greater than 60% of the standard extract from the user. By auditing the historical data of these credit card users, it can make bank bear the lowest risk in the future.