A Deep Multi-architectural Approach for Online Social Network Intrusion Detection System

The growth in complexity and danger in Modern cyber-attacks has necessitated developing strong, integrated, and adaptable intelligent defense systems. The need for increasing accuracy and lowering the necessary degree of human involvement, particularly feature selection during detection, are still unresolved problems. Thus, the most critical and relevant collection of features is critical for enhancing the effectiveness of intrusion detection systems. Another shortcoming of intrusion detection is its inability to adapt to changing network circumstances. Machine and deep learning approaches are now being used to solve the challenges mentioned above. This study attempts to address the issue mentioned above by combining an unsupervised deep learning technique with a heuristic way of class separation, proceeded by an upgraded Convolutional neural network for feature selection for classification. We detailed our suggested approach for learning features unsupervised, which is designed to minimize the amount of human expert involvement needed during the feature selection process. We suggest using deep learning to extract and choose required features from OSN users' behaviors before classification is done. We tested our suggested classifier on the NSL-KDD datasets and implemented it in a Weka application that supports GPU acceleration. Our suggested method shows a 99.89% accuracy that an unsupervised learning strategy is the best for detecting compromised accounts.