Securing the Internet of Things: Deep Learning Driven Intrusion Detection With Missing Data Imputation

Abstract

The Internet of Things (IoT), characterized by extensive connectivity, constantly generates and exchanges data among various devices. Ensuring the security of this data presents a critical challenge for IoT systems. Due to the inherent low-power nature of IoT devices and their minimal computing requirements, sensor usage patterns are heavily influenced by daily routines, leading to misclassification of normal patterns as intrusion attacks. This results in a high False Positive Rate (FPR) in the existing Intrusion Detection System (IDS). The reliance of these systems on weak features exacerbates the issue. Additionally, prevalent benchmark datasets lack missing values (removed in the pre-processing stage), thus reducing model robustness during training, which results in an elevated FPR. In response to these challenges, we propose a feature robustness framework that addresses missing values and enhances the feature set of the University of New South Wales-Network-Based 2015 (UNSW-NB15) dataset. This paper presents an improved framework for predicting missing values and estimating their gaps in real-time communication systems. The framework predicts and reconstructs missing values in real-time communication using an Imputed Missing Value-Multi Convolutional Neural Network (IMV-MCNN) for intrusion detection. The proposed system effectively detects and eliminates malicious devices and packets attempting to breach the network using the Grey Relational Analysis (GRA) for context security. Moreover, the proposed model illustrates that augmenting feature clusters in UNSW-NB15 improves accuracy and reduces the FPR. Experimental results demonstrate that the proposed model significantly improves detection performance, achieving an accuracy of 91.33%, increasing the detection rate by 6.33%, and reducing the FPR to 8.45%, thereby validating its effectiveness in securing IoT systems.

Topics & Concepts