The Application of Deep Learning Techniques in Computer Security for Big Data Anomaly Detection

Today, with the rapid development of the Internet, the requirements for network security are increasing day by day. This article explores the application of deep learning technology, especially convolutional neural networks (CNN), in anomaly detection in big data environments. The study designed a CNN-based anomaly detection framework to simulate the way the human brain processes visual information, achieving automatic extraction of complex features in data and recognition of abnormal patterns. Research methods include building multi-layer CNN models, data preprocessing, feature extraction and anomaly scoring, as well as comparative analysis of experimental designs. Experimental results show that CNN performs well in different network attack scenarios, with an average recall rate of 94.9%, a precision rate of 92.6%, and an average accuracy rate of 93.75%, which is significantly better than the local anomaly factor (LOF) and principal component analysis (PCA). ) and kernel density estimation (KDE) and other traditional methods. Additionally, CNN has lower response times when detecting advanced persistent threats, ransomware attacks, and distributed denial-of-service attacks. As an emerging abnormal behavior identification method, CNN can quickly and accurately find targets for users, which is of great significance to information security in today's Internet environment.