Securing Cloud‐Based Wireless Sensor Networks Against Botnet Attacks Using Multi‐Relational Cascaded Graph Convolutional Neural Network With Puma Optimizer

ABSTRACT The fundamental challenge is the precise identification of advanced attack patterns in a massive‐scale, heterogeneous traffic, maintaining data integrity and traceability. Inspired by these constraints, this work proposes a new Blockchain‐based Multi‐relational Cascaded Graph Convolutional Neural Network with Puma Optimizer (M‐CGCNNet‐PO). The process starts with three benchmark datasets: data collection N‐BaIoT, TON_IoT, and Bot‐IoT, followed by pre‐processing with a New Adjusted Min‐Max technique coupled with Decimal Scaling and Statistical Column Normalization. Deep semantic features are obtained through a Contrastive Transformer Network (CTNet), and subsequently input to the M‐CGCNNet for graph modeling in multiple relations. The Puma Optimizer optimizes the training, and intrusion logs are stored safely through a Blockchain‐based Proof of Evolutionary Model (PoEM). Experimental results demonstrate outstanding performance with all datasets, reaching up to 99.68% accuracy, 99.45% F1‐score, and high MCC and Kappa values, validating the robustness and reliability of the model. In summary, the suggested M‐CGCNNet‐PO provides a scalable, precise, and secure model for botnet detection in cloud‐based WSNs.