Improving IoT Security using Lightweight Based Deep Learning Protection Model

Abstract

The Internet of Things (IoT) has recently become an essential ingredient of human life. The main critical challenges that confront IoT are security and protection. Several methods have been developed to protect the IoT; among these methods is Intrusion Detection System (IDS) Deep Learning-based. On the other hand, these types of IDS have a complex operation that takes a long time when applied on IoT devices and is inconvenient for a massive system that includes many connected devices. Thus, this paper suggested a Lightweight Intrusion Detection System (LIDS) IoT model that depends on deep learning using a Multi-Layer Perceptron (MLP) network. LIDS has the following characteristics lightweight, high accuracy, high speed in detection, and deals with a few features in MQTT protocol. The MQTTset dataset was used in training, validating, and testing the proposed model to investigate the performance of the proposed LIDS. The achieved performance ratios for the proposed LIDS, as measured by accuracy and F1-score. The experiment results showed that for the balanced MQTTset dataset, the number of obtained features was 15 with accuracy (95.06) and F1_score (95.31). Also, for the imbalanced MQTTset, the number of obtained features was 12 with accuracy (96.97) and F1-score (98.24). The obtained results have shown the deep learning efficiency role in improving the accuracy of an intrusion detection model by approximately 3.5% compared to other methods in the literature. In addition, the proposed methods reduced the number of features by around 50% of the total number of features, leading to a LIDS operating in a constrained environment.