Abstract:

Abstract The Internet of Things (IoT) has transformed our world by offering enhanced accessibility, connectivity, and convenience in our daily lives. It facilitates the seamless flow of vast amounts of data among interconnected devices, creating a network that is susceptible to diverse network attacks and intrusions. Developing an efficient IDS (Intrusion Detection System) for IoT networks is a challenging task primarily due to two reasons: the massive amount of aggregated data and the diverse nature of IoT devices. Traditional IDS approaches struggle to handle and analyze this data in real time. Hence, there is a growing demand for advanced IDS techniques that leverage ML or DL methods. This study specifically focuses on intrusion detection in IoT networks, utilizing the UNSW-NB15 dataset. The UNSW-NB15 dataset is a well-known and publicly available dataset that is widely used for evaluating the effectiveness of IDS algorithms. The main purpose of the current work is to enhance the performance of intrusion detection by integrating feature extraction techniques based on genetic algorithms (GA) and ensemble machine learning algorithms (EM’s). By leveraging these approaches, the study aims to improve the accuracy and effectiveness of detecting intrusions in IoT networks. Feature extraction is a crucial step in IDS, as it aims to reduce the dimensionality of the dataset while retaining relevant information. Genetic algorithms, known for their optimization capabilities, are employed to search for an optimal subset of features that maximize the discriminatory power of the IDS. To achieve this, a framework is proposed that integrates genetic algorithms with various ensemble ML techniques, including random forests, Extra-Trees, XGBoost, AdaBoost, and stacking. The GA selects a subset of features from the UNSW-NB15 dataset, and the ensemble ML models are trained and evaluated using these selected features and calculate accuracy.

Keywords: