Big data mining algorithm for malicious behavior feature extraction and application in network security

The rapid escalation of cyber threats has necessitated advanced techniques for detecting and mitigating malicious behaviors in network environments. Traditional intrusion detection systems often struggle with large-scale datasets and complex attack patterns due to their reliance on manually crafted features and limited adaptability. To address these challenges, this study proposes a novel big data mining algorithm for malicious behavior feature extraction, specifically designed for largescale cybersecurity datasets. The algorithm integrates graph-based feature extraction and attention mechanisms to dynamically capture intricate relationships between network entities and prioritize critical features. Graph Convolutional Networks (GCNs) are employed to model complex interdependencies, while an attention mechanism assigns adaptive weights to features, enhancing the model’s focus on significant patterns. Additionally, the method leverages distributed computing frameworks to ensure scalability and efficiency when processing massive datasets. Experimental evaluations on CICIDS 2017 and UNSW-NB15 datasets demonstrate the algorithm’s superior detection accuracy, runtime efficiency, and scalability compared to state-of-the-art methods. The proposed approach achieves detection accuracies of 98.3% and 96.7%, respectively, outperforming baseline models while maintaining computational efficiency. By effectively balancing precision, recall, and runtime, this algorithm offers a robust and practical solution for modern cybersecurity challenges.