Enhanced IoT Security and Intrusion Detection Using Optimized Hybrid Deep Learning and Advanced Encryption Techniques

Abstract

This study introduces a hybrid deep learning architecture intended to improve intrusion detection and data security on the Internet of Things (IoT) platform. The model integrates Hierarchical Convolutional Neural Networks (HCNN) with a Calibrated Random Forest (RF) classifier, leveraging the spatial feature extraction and structured decision-making power. For better accuracy and lower complexity, the Enhanced Harris Hawks Optimization Algorithm (EHOA) is employed for feature selection. The system handles IoT network traffic via K-Means Clustering, label encoding, and normalization for proper input preparation for model training. Evaluation is done through principal metrics such as accuracy, precision, recall, F1-score, ROC-AUC, and PR-AUC. The model attains an accuracy of 83.41%, a precision of 85.29%, a recall of 74.3%, and an F1-score of 79.42%, with a high capability for detection even on imbalanced datasets. Furthermore, a KH-AES (AES-128) encryption mechanism is incorporated to protect data transfer, transforming sensitive traffic into ciphertext encryption while maintaining confidentiality and integrity. The hybrid method provides real-time and scalable anomaly detection while keeping false positives at a low level. With the integration of sophisticated feature selection, deep learning, and encryption, the system provides an efficient solution for managing the intricate security issues found in contemporary IoT networks.