Cybersecurity in the Internet of Things: the Detection of the Types of Upcoming Digital Information by Using Classification Techniques

This study addresses the critical challenge of Cyber-attacks detection (CAD) in the Internet of Things (IoT) environment, specifically focusing on the classification of non malicious and malicious network traffic. The primary objective is to enhance the accuracy and reliability of detection mechanisms through the implementation of advanced machine learning models, particularly the hybrid CNN-GRU-LSTM model. The study utilizes the SYN DoS dataset from the Kitsune Network Attack Dataset to train and evaluate various models, including Linear Discriminant Analysis (LDA), Logistic Regression, and the CNN-GRU-LSTM model. The methodology includes a comprehensive performance analysis of each model, employing metrics such as accuracy, precision, recall, and F1-score. The results reveal that both LDA and Logistic Regression achieved perfect accuracy (1.00), while the CNN-GRU-LSTM model exhibited an accuracy of 0.998. Additionally, the CNN-GRU-LSTM model demonstrated a high area under the curve (AUC) value of 0.8559, indicating strong discriminatory power. The study further employs SHAP (SHapley Additive exPlanations) for model interpretability, allowing for a detailed analysis of feature importance and insights into model behavior. In conclusion, the hybrid CNN-GRU-LSTM model offers a promising approach for effective network attack detection while providing a basis for future improvements in real-time applications and the exploration of additional datasets.