Advanced Multi Class Cyber Security Attack Classification in IoT Based Wireless Sensor Networks Using Context Aware Depthwise Separable Convolutional Neural Network
Bangar Raju Cherukuri · Journal of Machine and Computing · 2025
One of the most widely used wireless technologies in recent years has been wireless sensor networks (WSN), which has led to intriguing new Internet of Things (IoT) applications. Internet Protocol IP integration with IoT-based WSN enables any physical item with sensors must have widespread connectivity and transmit data in real time to the server linked to the gate on the internet. WSN security is still a developing area of study that falls under the Internet of Things paradigm. To protect digital infrastructures, strong techniques for precise and effective multi-class classification are required due to the growing frequency and sophistication of cyber-attacks. The proposed method makes use of the CICIDS2017 and UNSW-NB15 datasets alongside IoT-based wireless sensor networks to enhance cyber-security detection. In this work, Boosted Sooty Tern Optimization (BSTO) and Context-Aware Depthwise Separable onvolutional Neural Networks (CA-DSCNN) present an enhanced method for classifying multi-class cyber-security attacks. To guarantee consistent feature scaling, the proposed approach starts by applying Min-Max Scaler Normalization to preprocess the raw attack data. There is a feature selection stage that comes afterwards that uses Banyan Tree Growth Optimization (BTGO) combined with Augmented Snake Optimizer (ASO) to efficiently find and choose the most relevant characteristics to improve classification performance. Because of its strong feature extraction capabilities and computational efficiency, the CA-DSCNN is used; depthwise separable convolutions are used to strike a compromise between processing needs and accuracy. This architecture enhances the ability to extract complicated characteristics from the data and to comprehend those characteristics in context. BSTO is used to optimize the neural network's parameters, improving classification efficiency and accuracy in order to further enhance model performance. By lowering computational expenses and over-fitting, the proposed methodology which integrates IoT-based wireless sensor networks enhances cyber-security attack classification, exhibiting improved accuracy 99.5% and high PDR 99%.