Deep Learning with Multiple Faces to Improve Intrusion Detection in Adaptive Internet of Things Networks Optimization

Abstract IoT net security can be improved, and cyber threats may be reduced by using Deep Learning (DL) techniques, which offer a potential method for effectively detecting defects in network data. In this paper, DL techniques are utilized to build an improved IDS in IoT platform. Initially, a pre-processing phase is employed to handle the missing values and to identify anomalous data points via MissForest and Local Outlier Factor (LOF). Besides, a ResNet-50 approach is employed to extract specific and statistical features in the IoT data. Once feature extraction is done, feature selection is carried out using Improved Mutual Information (MI) method. Then, the dimensionality issues are reduced by Locally Linear Embedding (LLE) and an AdaptNet is introduced for detecting IoT attack using the combination of Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), and Auto-Encoder (AE). By leveraging advanced DL techniques and methodologies across different stages of IDS, the expected outcome is a robust and efficient tool capable of effectively safeguarding IoT networks. Use, AVOA and ARO optimization for fine-tuning pre-trained models on large datasets. Python tool is used for implementing the proposed work and the accuracy range is 99.2%.