Securing IoT Systems Using Artificial Intelligence-Driven Approaches

Abstract

The Internet of Things (IoT) has transformed modern infrastructure by connecting billions of smart devices, yet faces critical security challenges due to restricted processing power, diverse communication protocols, and delayed security implementations. Traditional cybersecurity approaches and conventional deep learning methods inadequately address these threats while maintaining computational efficiency for resource-constrained IoT environments. This paper presents a novel hybrid framework combining Discrete Orthogonal Hahn Moments with EfficientNet deep learning architecture for enhanced IoT attack detection. The methodology leverages Hahn Moments' superior feature extraction through weighted orthogonality properties to reduce dimensionality while preserving discriminative information. Integration with EfficientNet-B0's compound scaling and Mobile Inverted Bottleneck Convolution blocks enables efficient learning with only 5.3 million parameters a 77% reduction compared to traditional networks. Experimental validation demonstrates remarkable performance, achieving 99.6% detection accuracy with 99.63% specificity and 98.99% sensitivity at 232×232 resolution. The proposed framework outperforms K-nearest network (84.6%), Multiple Linear Regression (88.2%), Parse Tree (93.7%), Latent Semantic Analysis (97.9%), and conventional Deep Neural Networks (98%) while maintaining minimal computational overhead of 38 seconds. Results establish this hybrid approach as a robust solution for real-time IoT security monitoring in resource-constrained environments.