Network security vulnerability detection and repair model based on deep neural networks

Due to the growing complexity of network infrastructures, this investigation intends to create a Deep Learning (DL)-based model for identifying and fixing network security issues in digital infrastructures. Using the CIC-IDS-2017 network security dataset obtained and preprocessed through Forward Fill (FFill) to remove noise and determine the missing values, along with Z-score normalization for data normalization, provides the compatibility network layer. Local Binary Patterns (LBP) is used for the feature extraction stage that transforms structured data into concise and distinctive patterns to detect security threats effectively. Artificial Gorilla Troops Optimizer-driven Malleable Deep Neural Network (AGTO-MDNN) architecture is proposed to accurately determine the patterns associated with various security flaws. The model categorizes the vulnerabilities according to categories and severity levels, which allows for more focused responses. An integrated repair mechanism automatically suggests mitigations like software patches, configuration changes, or policy enforcements. Experimental evaluations establish that the proposed AGTO-MDNN model achieves superior detection performance by providing an accuracy of 97.32%, a precision of 95.27%, a recall of 96.58%, and an F1-score of 95.92% in comparison with the traditional DL-based models. An integrated repair mechanism improves response efficiency and reduces human intervention in the vulnerability management lifecycle. This research proposes a scalable and intelligent solution for proactive cybersecurity management and sets the foundation for integrating Machine Learning (ML) models in real-time vulnerability defense systems.