Web attack detection using deep learning models

Abstract

Due to the network access and security vulnerabilities of web applications, web applications are often targets of cyber-attacks. Attacks against web applications can be extremely dangerous. A lot of damage has been done because of the vulnerability of the application, which lets them access the Web Application database. Monitoring web attacks and generating alarms when a challenge to an attack is detected. This work uses deep learning models (ANN, CNN & RNN) to detect web attacks automatically. To identify the time when the attack on the payload occurred, the work first analyses the web log information provided by the user. To make an attack prediction, the log information is pre-processed. Web-log information is pre-processed to remove duplicate values and missing values and to get the payload information. To encode the fields and normalize (Min-Max) that converts into unique format while predicting and the encoding value also applied. To construct the prediction model for the detection of web attacks, the pre-processed dataset is incorporated into the deep learning classifiers. In the performance evaluation, RNN provided 94% accuracy and 6% error rate, higher than other method.