Network Intrusion Detection: A Study on Various Learning Approaches

Abstract

Internet of Things (IoT) is a comprehensive paradigm where millions of devices are connected to a network. These interconnected devices create a network of intelligent systems that exchange data without the need for any computer or human communication. The devices gather data that is important to humans and businesses. Standard high-end security solutions are ineffective for safeguarding an IoT system because IoT devices have limited storage and processing capability. Due to the proliferation of innovative attacks, network security is finding it difficult to identify breaches with good accuracy. As a result, it becomes necessary to provide smart security solutions that are portable, widely dispersed, and provide long term services. The monitoring of network traffic by an intrusion detection system (IDS), which protects against prospective intrusions and preserves the network's confidentiality, integrity, and availability, is one solution. However, IDS still has difficulties detecting intrusions and improving detection accuracy while lowering false alarm rates. When dealing with heterogeneous data of varied sizes, Machine Learning (ML) and Deep Learning (DL) have already demonstrated their importance. Many modern IDS are ML based models. In this paper, ML and DL learning models like Random Forest (RF), Decision Tree (DT), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), XG Boost (XGB), Multi-Layer Perceptron (MLP), Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) are used and compared. The best algorithm amongst these is compared with the existing state-of-art models. The dataset used is UNSW-NB 15 Train set and Test set. The metrics used for comparison are Accuracy (Ac), Recall (Rc), Precision (Pr), F1 score, Mean Squared Error (MSE), training time, prediction time and total time. RF performs better than all other algorithms with Train Ac of 95.98% and Test Ac of 97.69%. It also outperforms the existing state-of-art models achieving the highest accuracy.