DEEP LEARNING BASED MALICIOUS TRAFFIC DETECTING IN IOT NETWORK

Abstract

In order to identify and categorize network traffic, we developed an Intrusion Detection System (IDS) model in this study that is based on supervised machine learning techniques. We prepared the dataset for supervised learning techniques. The type of data resolution, imbalance, and missing values are the main characteristics of the preparation of the dataset. The dataset is presented in terms of network layers and the characteristic selection uses Random Forest (RF) form. After that, the network and transport cluster is created. To develop a four-model, four machine learning algorithms are applied to the provided dataset, which consists of all lay ers. The model was constructed with the aid of Artificial Neural Network (ANN), RF, Support Vector Machine (SVM), and Naive Bayes (NB). We conducted the studies using multi-class classification, layer clusters, complete features, and significant features. The evaluation of the experiments is important to see the accuracy, processing time, and parameter tuning to calculate the production stage performance of the algorithm. The findings show that Random Forest is the most effective choice for adjusting accuracy and efficiency, particularly in tasks centered on network communication; in addition to being more diverse and achieving the highest accuracy when compared to state-of-the-art, it also modifies the IoT-23 dataset to lower computational complexity and resource consumption. It is anticipated that this study will further the field of IoT network intrusion detection. The achieved accuracy depends on different algorithms and clusters and is from 58 to 78; with highest accuracy Random Forest has been suggested. The time taken by classification is also s hown; with the lowest t ime, Naive Bayes is not only the most efficient process but also the fastest.