Integrating Dung Beetle Optimization in Machine Learning for Advanced Zero-Day Attack Detection and Classification

Integrating Dung Beetle Optimization in Machine Learning for Advanced Zero-Day Attack Detection and Classification

연구 분야: Strategies

학회: 2024 9th International Conference on Communication and Electronics Systems (ICCES)

With the widespread adoption of Internet such as online banking, health systems, e-commerce, and other daily requirements, the risk of becoming revealed to numerous is growing proportionally. Zero-day Attacks targeting unidentified vulnerabilities of a system or software start-up additionally investigate direction in the cyber-attacks fields. Recent techniques both apply machine learning (ML) and anomaly-based approaches to protect against this attack. In the ML umbrella are traditional ML-based models that are familiar with having low detection rates and prediction quality about data that it cannot yet be proficient in. DL-based approaches, particularly convolutional neural networks (CNNs) with regularization techniques, tackle these problems provide a superior prediction quality with unidentified data, and avoid overfitting. Identifying zero-day attacks over this technique miscues various parameters including the frequency of certain byte streams in network traffic and their correlation. This study designs a new Machine Learning with a Dung Beetle Optimization Algorithm using the Zero-Day Attack Detection and Classification (MLDBO-ZDADC) technique. The aim of the MLDBO-ZDADC method is in the automated and effectual detection of zero-day attacks. At first, the MLDBO-ZDADC approach applies min-max normalization to pre-process the input data. Besides, the MLDBO-ZDADC technique exploits the feed-forward neural network (FNN) model for attack detection and its parameter tuning process is performed by using the dung beetle optimization (DBO) model. To ensure better prediction results of the MLDBO-ZDADC approach, a sequence of experiments is applied to benchmark dataset. The experimental validation illustrates the promising accomplishment of the MLDBO-ZDADC approach on the existing methods.