Nesnelerin interneti öğelerinin iletişiminde güvenlik tedbirlerinin artırılması

The development of information technologies makes significant contributions to various aspects of our life. It brings various innovations together, depending on the rate of technology's growth. The Internet of Things (IoT) is among the most popular and fastest expanding technologies in recent years. Addressable IoT devices generate and use significant data over the Internet. Because of the increase in data traffic, attacks in IoT networks are also increasing significantly. The present thesis increases security in IoT communication by providing binary and multi-label classification methods for identifying attacks in IoT networks. A new Hybrid Deep Learning model has been designed for detecting intrusions. Two different public datasets (CIC-IDS-2018, BoT-IoT) are used for the proposed Intrusion Detection System (IDS). In addition, a new dataset containing routing attacks targeting IoT devices has been created, because of the increase in routing attacks in IoT networks in recent years. Hence, 14 attack types have been analyzed in the present work. The results of the analysis have been presented extensively, as to the accuracy, F1-score, and training time of the model. Comparing our results to previous studies showed that our designed hybrid Deep Learning model has the best training time/accuracy and time/F1-score performance ratios. Comparisons prove that our proposed model is more successful in detecting IoT attacks than the previous works. The successful performance of our proposed model is proof that hybrid Deep Learning methods can be an innovative and efficient perspective in IoT Intrusion Detection Systems.