Ensemble-Guard IoT: A Lightweight Ensemble Model for Real-Time Attack Detection on Imbalanced Dataset

Abstract

The rapid increase in the number of IoT devices has made ensuring robust real-time attack detection more critical than ever. The volume of data being accessed in real-time by these devices presents unique security challenges that traditional detection techniques struggle to address with the required precision and efficiency. To overcome these limitations, we have developed Ensemble-Guard IoT; an innovative ensemble model combining Gaussian Naive Bayes (GNB), Logistic Regression (LR) and Random Forest (RF) through soft voting classifiers. Ensemble learning by combining multiple machine learning models offers a significant advantage in reducing computational costs compared to deep learning models, making it a practical solution for real-time applications. We performed a thorough evaluation of our proposed scheme in terms of accuracy 99.63%, precision1.00%, recall 99%, f1-score 1.00% and computation time 524.40s. We also compared the performance of our scheme with the classical schemes. Our comprehensive evaluation demonstrate that Ensemble-Guard achieves highest average accuracy of 99.63% thus validating the effectiveness of our scheme in identifying IoT attacks in real time. This hybrid voting system combines the predictions from different classifiers, ensuring a more balanced and accurate final decision. Ensemble-Guard IoT is a significant step forward in safeguarding IoT infrastructures, offering a scalable and cost-effective solution to the evolving threat landscape.

Keywords

• Ensemble-guard IoT
• ensemble learning
• cybersecurity
• machine learning and real time attack detection