Generative Adversarial Network for Enhancement Network Security Log Detection

Our study innovates in network security by preprocessing heterogeneous log data to eliminate unnecessary elements, ensuring uniformity post-conversion, and amalgamating data using temporal and associative techniques. We address data imbalance with an advanced Seq-GAN, generating minority class samples to enrich the dataset. Furthermore, we extract semantic vectors from log data, resulting in 360,899 high-quality attack entries, and transform log IP addresses into a continuous feature space for improved threat detection. Our approach, leveraging adversarial augmentation and natural language processing, uniquely identifies malicious web entities and enhances log data analysis for threat detection.