Optimizing Cybersecurity Against Zero-Day Attacks Using Dandelion Optimization and Machine Learning Techniques
J. Vanitha, P. Anandababu · 2025
Numerous Intrusion Detection and Prevention Systems (IDPS) have been presented to detect mistrustful actions. However, attackers are using novel susceptibilities in methods and more progressive cyber-attacks; the zero-day attack stays unseen from IDPS in maximum instances. This feature has inspired several investigators to present diverse Artificial Intelligence (AI)-based models to respond, detect, and prevent such developed attacks. Machine Learning (ML) techniques effectively categorize data samples into their corresponding classes. The typical ML assessment method assumes testing data samples from pre-seen categories applied within the training stage. ML-based NIDS encounter a novel attack traffic named zero-day attack, which is not applied in training owing to their nonexistence over time. In this work, a new Optimizing Cybersecurity against Zero-Day Attacks using the Dandelion Optimization Algorithm and Machine Learning (OCZDA-DOAML) method has been proposed. The objective of the OCZDA-DOAML method is to utilize hyperparameter-tuned ML models to detect cybersecurity against Zero-Day Attacks. The OCZDA-DOAML technique employs linear scaling normalization to transform the raw data into a standardized format. The Kernel Extreme Learning Machine (KELM) technique is employed in the zero-day attack detection process. Finally, the Dandelion Optimization Algorithm is used to optimize the hyperparameter tuning of the KELM technique. The simulation process of the OCZDA-DOAML technique is accomplished under ToN-IoT and CIC-IDS-2017 datasets. The experimental assessment of the OCZDA-DOAML approach illustrates a superior accuracy value of 95.96% over the existing methods in terms of various analysis parameters.