A Sophisticated Cybersecurity Intrusion Identification Model Using Deep Learning

Abstract

The data is vulnerable to numerous attacks during transmission within the network environment. Identifying breaches
in network communications is becoming more critical. Scientists employ machine learning methodologies to develop
efficient Intrusion Detection Systems (IDS). This paper presents an IDS incorporating preprocessing techniques and
a Deep Learning (DL) framework for detecting Denial of Service (DoS) attacks. The research evaluated the proposed
model utilizing the dataset commonly referenced in academic literature. The study implemented preprocessing
techniques, including feature deletion, random subset choice, choosing features, duplication elimination, and
normalizing on the database. Enhanced recognition efficiency was achieved for both training and testing assessments.
The test results indicate that the Conventional Neural Network (CNN)-based inception-like model had the highest
accuracy, with 98% for binary classification and 97% for multiclass classification among the offered models. The
deductive time of the suggested framework for different test data sizes appears favorable compared to baseline models
with fewer parameters that can be trained. The proposed IDS system, along with the preprocessing techniques, yields
better outcomes when compared with contemporary studies.