Abstract:
The widespread use of the internet has brought immense convenience, but it has also led to a
rise in cyber crimes. Attackers are using various tactics and techniques to compromise the
security of information systems. One of the major threats in this landscape is web attacks,
which pose a serious threat to web applications. Extensive work has been done for web security
through multiple detection and prevention tools at each layer of security. Tools like IDS, IPS,
and SIEM solutions have been proposed to detect and prevent these attacks. These security
solutions mainly rely on network traffic stats (flows), signatures, cyber threat intelligence
(CTI), and static threat detection rules. These methods have protected web security, but there
are some limitations observed toward advanced attack payloads that use sophisticated
techniques, a limited number of attempts, and zero-day exploits. This research aims to identify
malicious web traffic using an innovative approach that combines deep learning with spatial
pyramid pooling (SPP) to detect attacks on the base of payloads in network traffic. Deep
learning is a powerful tool for recognizing patterns and extracting features from images. The
proposed method involves using image classification techniques to dynamically spot different
types of web attacks on the fly. By converting both malicious and clean payloads into image
formats, the model has been trained to classify these data into either malicious or clean
categories. Additionally, SPP techniques have been used to adapt the model to varying sizes of
images. This method will help to improve the efficiency of the model by avoiding information
loss due to resizing and cropping images to a fixed size. This work automates the process of
extracting meaningful features, eliminating the need for manual feature selection commonly
used in traditional machine learning approaches. The proposed approach aims to provide a
more effective defense against evolving web attacks.
