Implementing malware detection framework using adaptive and deformable attention-based elman residual recurrent...

The rapid expansion of the Internet of Things (IoT) has developed various security vulnerabilities, as conventional defense mechanisms often struggle based on heterogeneity and resource constraints of smart devices. This research aims in designing a robust and high-performance security framework, which integrates an advanced deep learning with optimized cryptography to protect IoT networks from malware threats. The proposed methodology employs a novel Adaptive and Deformable Attention-based Elman Residual Recurrent Neural Network (ADA-El-RRNN) for precise malware detection, whereas the hyperparameters are fine-tuned by using Opposition Humboldt Squid Optimization (OHSO) to prevent local optima and premature convergence. Following detection, secure data transmission is ensured through Optimal Key-based Elliptic Curve Cryptography (OKECC), which utilizes OHSO for efficient key generation to minimize computational overhead and memory usage. The experimental results across four standard datasets demonstrate the framework's superior performance, achieving peak malware detection accuracies up to 97.4% and a specificity of 98.5%, significantly outperforming existing models like LSTM, RNN and standard ECC. This work is significant as it provides a scalable, low-latency solution that enhance the data confidentiality and integrity, effectively safeguarding sensitive information in dynamic real-world IoT environments.