Аннотация
This paper presents a comprehensive investigation into the implementation of cutting-edge machine learning paradigms for fortifying cybersecurity on the Internet of Things (IoT) ecosystem, with a particular focus on two pivotal datasets: the extensively utilized IoT-23 dataset and the specialized CRAFTED—Cooja RPL Attack Framework Test and Evaluation Dataset. Employing empirical analyses of simulated cyber threats, this research constructs multiple machine learning models to predict and classify a wide spectrum of IoT attack vectors and parameters with enhanced precision. Through rigorous experimentation, we evaluate an array of deep learning architectures, including Long Short-Term Memory (LSTM) networks and Residual Networks (ResNet), juxtaposing their efficacy against conventional machine learning models. The results showed that deep learning methods are good at identifying complex patterns in multidimensional data. However, traditional models such as Random Forest (RF) and Decision Trees (DT) have proven to be more efficient in terms of processing speed and classification accuracy in real time, making them the preferred choice for practical applications. The study also used methods for selecting features and adjusting model parameters, which significantly improved their accuracy. These approaches have provided useful technical insights and solutions for timely threat detection in IoT networks. The work carried out creates the basis for further research in the field of identifying cyber threats and anomalies, contributing to the development of reliable and intelligent cybersecurity systems.