A Robust System for Detecting and Preventing Payloads Attacks on Web-Applications Using Recurrent Neural Network (RNN)

Abstract

Due to the growing rate of Internet usage, web apps have become the most popular Internet application. This has made web applications a significant objective for cyber-criminals; thereby, carrying various attacks on web applications like Cross-Site Scripting (XSS), Structured Query Language Injection and Shell attacks. Because of the high rate of web-based assault, this paper presents a robust framework for detecting and preventing multiple payload attacks on web applications. In this paper, an RNN model was trained on a dataset that contains different categories of assaults that are carried out on web applications. These attacks include: XSS, SQLi, and Shell. Random Over Sampling approach was used to resolve the problem of highly imbalanced dataset which prepared the dataset for preprocessing. After solving the imbalanced problem, pre-processing was then carried out on the dataset by performing data cleaning and tokenization. The tokenized data was transformed into an array which was used in feeding our RNN model as input. Our proposed model was trained on two (2) epochs, where each of the epochs shows the accuracy and loss values obtained by the model for both training and testing data. After training, our proposed RNN model gave us an accuracy of 99.96% for testing data and 99.91% for training data. We also deployed our RNN model to the web by making use of a python flask to build a robust system for detecting and preventing different payload attacks on web applications. This paper is limited to web-application attacks.