A novel deep learning-based intrusion detection system for internet of things networks

The rapid advancement of technology today has increased the opportunities provided by internet-based services is significantly simplifying our daily lives. The development of Internet of Things (IoT) technology has particularly enabled various devices to communicate with each other and with users via the internet, simplifying access to different services. While IoT refers to the integration of physical objects with the virtual world, it allows these objects to interact with each other or with larger networks via the internet. The widespread use of this technology has transformed areas such as healthcare, agriculture, energy, transportation, and many others, finding applications ranging from smart cities to home automation systems. The most common types of attacks on IoT networks include denial of service (DoS), data leakage, identity theft, and malware attacks. Attackers often exploit software vulnerabilities, weak password practices, and network security gaps to carry out such attacks. For example, a botnet is a structure in which attackers remotely control a large number of compromised IoT devices to form a network. Through this structure, the devices can be used as part of malicious traffic and can carry out coordinated attacks that cause large-scale network disruptions. These security vulnerabilities in IoT environments can lead to serious threats, such as unauthorized access and data breaches. Particularly, the protection of personal data is critical to ensuring the security of IoT devices and networks. Security experts are constantly developing new methods to ensure IoT security and designing more effective security protocols and software to protect these devices. IoT security solutions include systems capable of monitoring network traffic, detecting threats, and responding quickly. These systems often utilize artificial intelligence and machine learning techniques to detect behaviors deviating from the norm and predict potential attacks in advance. Advanced detection systems can instantly identify attacks targeting IoT devices, thereby preventing damage to devices and networks. The advantages brought by IoT technology are nearly undeniable, however, to ensure the safe use of this technology, continuous innovations in the field of security and the strengthening of security measures are necessary. Newly developed security strategies and technologies play a crucial role in reducing threats to IoT devices and networks and ensuring that this new technological era is adopted safely. In this thesis, an intrusion detection system designed to operate in IoT networks has been developed. While modeling the system, deep learning algorithms have been extensively utilized. Long Short-Term Memory (LSTM) neural networks, Gated Recurrent Unit (GRU) neural networks, Convolutional Neural Networks (CNN), and Autoencoder (AE) Neural Networks are the main deep learning algorithms used. These algorithms have been used in a hybrid manner within the model. Data sets such as CIC-IDS-2018, CIC-IDS-2017, CIC-IOT-2023, UNSW-NB15, N-BAIOT, BOT-IOT, CIC-DDOS-2019, TON-IOT and IOT-23 which are frequently used in the literature in this field, have been employed as data sets. The data sets were initially preprocessed. Irrelevant data were removed, necessary data were converted into numerical formats, erroneous data were corrected, and missing data were completed using various methods. The data were normalized within a specific range. A model was designed that would be compatible with all data sets used and would yield good results. An average success rate of 98.7% has been achieved across all datasets with the model.