A Deep Learning Approaches with Optimal Padding Advanced Encryption Standard for Web Platform Security Monitoring and Secure Data Transmission

Abstract

High-dimensional network traffic, dynamic attack behaviour and complex spatio-temporal dependency impose severe demands on a web platform’s accurate, real-time security monitoring. Traditional machine learning and deep learning-based models like CNN-based optimization, LSTM networks, ZSMMS statistical models, RF2RFGB ensemble learning and Decision Tree classifiers have faced challenges in terms of inadequate feature representation, compromised generalization capability and limited robustness in the face of evolving cyber-attacks. To alleviate these challenges, this paper proposes a meta-centric hybrid deep learning framework for web security monitoring and attack detection. First, feature optimization is conducted based on advanced hybrid selection strategies to select the most discriminative behavioural and anomaly-based features from the CIC-MalMem-2022 dataset, which helps to reduce redundancy and improve convergence efficiency. Then the extracted optimal feature set is fed to Multi-Layer Perceptron (MLP) neural networks combined with ensemble learning methods like AdaBoost, XGBoost and Stacking to achieve accurate classification of normal and malicious traffic. Finally, an Optimal Padding Advanced Encryption Standard (OPAES) module is incorporated after classification to guarantee the secure transformation, encryption and transmission of detected results with enhanced confidentiality and integrity.