Enhancing Cybersecurity with Continual Spatio-Temporal Graph Convolutional Network for Early Detection of Malicious Activities and Improved Network Security

Hardware, software, data storage and applications are all connected by the Internet of Things (IoT) to continuously offer services to businesses. However, this integration also creates new potential entry points for cyberattacks. IoT device privacy is highly vulnerable to threats such as software piracy and viruses. These risks can compromise private information, resulting in significant financial losses and reputational harm to a company. In this paper, Enhancing Cybersecurity with Continual Spatio-temporal Graph Convolutional Network for Early Detection of Malicious Activities and Improved Network Security (EC-CSTGCN-DMS) is proposed. Initially, the input malware samples are gathered from the Malimg dataset. The input samples are given to Unsharp Structure Guided Filtering (USGF) for resizing images, pixel normalization and removal of noise and irrelevant details from the input image samples. The preprocessed samples are provided for feature extraction utilizing Component Separable Synchroextracting Transform (CSST) for extracting shape, edge and texture features. The feature vector that has been extracted is presented to the Continual Spatio-Temporal Graph Convolutional Network (CSTGCN) to detect malware. In this framework, the temporal component refers to structural dependency progression derived from spatially ordered malware image patches rather than chronological time-series analysis. To guarantee proper and efficient classification, the Educational Competition Optimizer (ECO) is used in order to optimize the network weight parameters in the CSTGCN. The proposed technique is implemented in Python. The efficiency of the proposed method is evaluated using several performance criteria, including F1-Score, Accuracy, False Alarm Rate (FAR), Detection Rate and Training Time. The proposed framework for enhancing cybersecurity in IoT environments achieved remarkable results. The proposed method achieved 98.45% accuracy, 97.82% F1-Score, 98.95% detection rate, 0.84% FAR and a training time of 48 min, outperforming existing approaches such as An Inception V3 approach for malware classification utilizing Machine Learning (ML) and transfer learning (MC-InceptionV3), Novel hybrid Deep Learning (DL)-driven Cyber Security Threat Detection Model with optimization algorithm (CSTD-ELSTM), and A novel DL-driven approach for malware detection (MD-SVM). This innovative approach effectively contributes to securing IoT networks against dynamic and evolving cyber threats.