Intrusion Detection System Using DL for Iot Devices

Abstract

The growth of Internet of Things (IoT) devices has significantly transformed various domains, including smart homes, healthcare, and industrial automation. However, the increasing connectivity also introduces substantial security vulnerabilities, making the development of effective Intrusion Detection Systems (IDS) essential , in this study we proposed an Intrusion Detection System framework using deep learning techniques to enhance security in IoT environments through binary classification. The proposed framework integrates several deep learning models, such as Autoencoder “AE”, One-Dimensional Convolutional Neural Network “CNN1D”, Two-Dimensional Convolutional Neural Network “CNN 2D”, Long Short-Term Memory “LSTM”, FeedForward Deep Neural Network “FFDNN” and “Vgg16” , “MobileNetV2” , DenseNet 121” combination Model, the data preprocessing involves normalizing datasets, feature extraction using a deep autoencoder, and balancing using the Synthetic Minority Over-sampling Technique “SMOTE”, the models were trained and evaluated using the CICIDS2017 and CICDDoS2019 datasets, which encompass a diverse range of network traffic behaviors and attack scenarios. Experimental results indicate that most models achieved high accuracy, but the critical factor for choosing the best model is the recall because in an IDS a low false positive rate is crucial as undetected attacks pose significant dangers, the best-performing model was combining Autoencoder and CNN 2D Model, achieved impressive results with an accuracy of 98,23%, recall of 96,82%, precision of 99,63% , and F1-score of 98,20% for CicIDS2017 Dataset and accuracy of 98,43%, recall of 99,38%, precision of 97,59% , and F1-score of 98,48% for CicIDDOS2019 Dataset , the integration of autoencoders with convolutional and recurrent neural networks provides a comprehensive feature representation, improving the system's ability to identify both known and novel threats. In conclusion this research demonstrates the efficacy of deep learning-based IDS in fortifying IoT security, The findings suggest that such systems can offer a reliable and efficient solution for real-time intrusion detection, paving the way for future advancements in IoT cybersecurity. This work aids on mitigating cyber threats, in increasingly interconnected environments, highlighting the potential of advanced deep learning methodologies in enhancing network defense mechanisms.