Enhancing Smart Camera Attack Classification with CGAN based Data Augmentation

Abstract

The Internet of Things (IoT) has transformed the way individuals interact with their environments, enabling automation and enhancing operational efficiency. However, the widespread adoption of IoT devices has raised significant concerns regarding privacy and cybersecurity risks. In this study, we create a custom smart cameras dataset and develop machine learning (ML) and deep learning (DL) models to classify and detect different attack types in smart camera network traffic. To address data imbalance, Conditional Generative Adversarial Networks (CGANs) were utilized to generate synthetic data. A comparative analysis of ML models, including Random Forest (RF), K Nearest Neighbors (KNN), and Support Vector Machine (SVM), and for DL models, we perform Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), and Attention DL models. The results show that DL models, particularly CNN, achieved exceptional classification accuracy, exceeding 94% for most attack types. Among the ML models, SVM with accuracy of 94.25% achieves higher performance than other ML models. This research contributes to enhancing smart cameras security by demonstrating the potential of conditioned synthetic dataset generation to improve the performance of advanced DL and ML techniques in identifying and mitigating cybersecurity threats, thus ensuring the protection of user data and privacy.