Research On Abnormal Behavior Prediction Of Information System Users Based On Deep Neural Network

With the continuous improvement of the level of enterprise information,the core business of the enterprise is increasingly dependent on the reliable operation of the information system.For enterprises,the abnormal operation of any information system user may bring immeasurable losses to the enterprise.Therefore,the negative impact of user abnormal behavior on the enterprise is a problem that cannot be ignored.The detection of user abnormal behavior of has become a research hotspot for scholars.In various user abnormal behavior detection methods,the data-driven method can adapt to the complex changes of the data,so the data-driven method can be used for user abnormal behavior detection.Due to the variety of data-driven methods,how to choose an efficient and accurate user anomaly prediction method has become one of the research questions in this paper.In addition,most of the anomaly detection software in the enterprise is developed for the software and hardware of the system.There are few system software that specifically detect user behavior,and the analysis process of user abnormal behavior takes a long time,and there are more steps to implement automated operations.Therefore,it is urgent to develop a system software for identifying user abnormal behaviors,which helps enterprises improve data processing efficiency and increase data value.In order to solve these two problems,this paper takes a shipbuilding enterprise as an example,adopts the theory and method of feature engineering,carries out feature processing on the information system log data,and carries out detailed analysis and experimental research on related classification models and algorithms.Finally,the prediction system software is designed based on this,and an identification and prediction framework for user anomaly behavior is extracted.The main research and work done in this paper is as follows:(1)Abnormal behavior classification and feature engineering construction of log data.This paper identifies the conditions under which anomaly behavior occurs,transforming unsupervised log data sets into a supervised user anomalous behavior data set.Then,using feature matching,construction,coding,scaling and dimensionality reduction methods,the log data is subjected to feature engineering processing,and the importance ranking results of the data set and feature dimension suitable for the deep neural network model are obtained.(2)Deep neural network model prediction.In this paper,the DNN model is constructed and optimized firstly.Secondly,the feature tail-off method is used to iteratively test the performance of the model with different feature dimensions,and the relatively good prediction results are obtained.Then,the different behavior patterns in each dimension are compared.Then this paper compares the incidence of different behavior patterns in each dimension,and visualizes the comparison results,and analyzes the causes,and develops strategies to reduce the incidence of abnormalities.Finally,the paper compares and analyzes the prediction results of deep neural network model and multiple linear regression algorithm and support vector machine algorithm,and expounds the superiority of deep neural network model.Taking the business department as an example,the predicted recall rate of deep neural network is 77.4%,which is better than 74.86% of support vector machine;the prediction accuracy is 84.56%,which is significantly better than 63.03% of support vector machine.(3)System design application and predictive framework refinement.Combined with relevant theoretical methods and actual experimental results,the paper designs the identification prediction system software based on this and extracts the identification and prediction framework of user anomaly behavior,which shortens the time of data processing analysis,and improves the efficiency of identification and prediction,and makes the analysis of the user abnormal behavior easier.