A Hybrid Approach of CNN and LSTM to Detect Intrusion in Edge IoT Devices using CatBoost
Md. Al Shahriar, Ashim Dey · 2023
Intrusion into IoT devices is an unauthorized action threatening the security, privacy, and functionality of interconnected devices. As the use of IoT devices grows, security concerns rise, with potential vulnerabilities targeted by attackers. So, it is very important to create a system that can detect malicious activities inside the network. In this work, the complication of detecting intrusion is addressed in IoT networks by utilizing the well-regarded Edge IIoT dataset which contains 14 attacks along with one normal class data. Throughout this study, we have implemented three different neural network models: CNN, LSTM, and a hybrid (CNN+LSTM). Before training these models, we have gone through rigorous data preprocessing steps. Firstly, the dataset contains 61 features, some of them are numerical, and others are categorical. The numerical values are easily computable, but the categorical values are not. So, CatBoost has been applied to encode this categorical value into numerical ones. Secondly, there is a lack of balance in data occurrence frequencies in different classes. To address this issue, oversampling and under-sampling are carried out. After that to enhance the efficiency and reduce the computation time of the models, important features were identified using the Information Gain method. Comparative analysis shows that all three models performed well on the final preprocessed dataset. Among them, the hybrid (CNN+LSTM) model exhibited superior accuracy, reaching 99.99%. We hope this study will strengthen cybersecurity by accurately identifying attacks in edge IoT devices.