Classification and Explanation of Different Internet of Things (IoT) Network Attacks using Machine Learning, Deep Learning and XAI

Abstract

The internet of things is one of today’s most revolutionary technologies. Because of its pervasiveness, increasing network connection capacity, and diversity of linked items, the internet of things (IoT) is adaptable and versatile. The most common problem impeding IoT growth is insufficient security measures. The threat of data breaches is always there since smart gadgets gather and transmit sensitive informa tion that, if disclosed, might have severe consequences. Modern advances in Artificial Intelligence are providing new Machine Learning and Deep Learning approaches to address more complex issues with greater model performance. This predictive capac ity, however, comes at the cost of growing complexity, which can make these models hard to understand and interpret. Though these models give highly precise results, an explanation is required in order to comprehend and accept the model’s decisions. Here comes XAI which emphasizes a variety of ways for breaking the black-box nature of Machine Learning and Deep Learning models as well as delivering human level explanations.In this article, to identify and classify IoT network attacks, we have analyzed six machine learning and deep learning approaches: Decision Tree, Random Forest, AdaBoost, XGBoost, ANN, and MLP. Accuracy, Precision, Recall, F1-Score, and Confusion Matrix are some of the metrics we have used to evaluate our models. We have achieved fairly impressive results (above 96%) in binary clas sification for all the techniques. When all of the classifiers were analyzed, Decision Tree and Random Forest outperformed all others (above 99%) for both binary and multiclass classification. Adaboost and ANN, on the other hand, perform badly for multiclass classification. We have also applied Undersampling, Oversampling, and SMOTE techniques on a dataset to reduce data skewness and to evaluate multiple ML and DL algorithms.We have used LIME, SHAP, and ELI5 approaches to inter pret and explain our models. The feasibility of the techniques suggested in this work is demonstrated in the IoT/IIoT dataset of TON_IoT datasets, which incorporate data obtained from telemetry datasets of IoT and IIoT sensors.