Robust Framework for Detecting and Classifying Network-Based Attacks Using Ensemble Machine Learning Approach

Cybersecurity threats are increasingly sophisticated, making it critical to design advanced systems for detecting and mitigating network-based attacks. However, existing approaches often face limitations in accuracy, scalability, and adaptability to diverse types of attacks. This research addresses these challenges by suggesting a robust model that combines advanced machine learning (ML) and optimization methods to deliver superior performance in detecting and classifying network-based attacks. The self-collected dataset of 7,000 instances with 26 features. The preprocessing stage includes duplicate record removal, missing value imputation, and feature normalization to ensure dataset consistency. Feature extraction incorporates statistical, network-based, and behavioral features, for identifying abnormal traffic patterns. A hybrid feature selection approach using the Zebra Optimization Algorithm (ZOA) and Pelican Optimization Algorithm (POA) is employed to enhance model performance for effective attack detection. This work introduces a network attack detection scheme utilizing Ensemble Machine Learning (EML) methods for enhanced detection of cybersecurity threats. Additionally, Prairie Dog Optimization (PDO), is used to optimize hyperparameters for superior model performance. The proposed framework demonstrates a high accuracy of 99.7%, recall of 98.4%, and precision of 99.4%. Thus, the outcomes establish the superior efficiency of the developed technique for efficient threat detection, offering a scalable and adaptable result for cybersecurity threats.