Research On SQL Injection Detection Technology Based On Machine Learning

With the development of computer technology in recent years,a wide range of Web applications have been merged to people's lives and business.While network bring great convenience to people,people's lives are increasingly inseparable from the network.However,the networked property and information also led to a variety of security issues.Among all of the network attack,SQL injection has the advantages of low difficulty,high frequency and fast technology update,and its main hazards include the leakage of user information,the access of system privileges,and the tampering of web information.At present,the main method to defense SQL injection including safe encoding process,detection systems based on program analysis and feature matching etc.Both of them have the problems of high information requirements,long development process,hard to deploy etc.Therefore,this thesis proposes a SQL injection detection technique which can be used in the HTTP application layer,and detect the potential threat only by analyzing user input.In this thesis,a SQL injection detection model is proposed,which is based on user input in the order of data pretreatment,lexical analysis,grammar analysis and machine learning.This SQL detection model has these characteristics :do not need other background information besides user input;recursive decode the user input to resolve the possible payload as far as possible;abstract the syntax tree of sample after lexical analysis and grammatical analysis before machine learning;has more reliable validation results with double machine learning verification;can distinguish the level of threat for users so that they can deal whit different levels of attack more convenient.The experimental results show that compared with other SQL injection detection models,this proposed model has high accuracy and efficiency in the case of only a small amount of information,and can effectively detect SQL injection attacks.