In this paper, due to limited memory, processing power, and lack of security features in many IoT devices, these devices are particularly vulnerable to network intrusion attacks. To address this problem, the paper stresses the importance of implementing effective security measures and demonstrating the potential of learning technologies to detect and prevent network intrusion attacks on these devices. The proposed framework is useful for enhancing device security and reducing the risk of attacks and data breaches. We used an intrusion detection system with artificial intelligence (AI) by using DNN and then tested with the dataset (KDD Cup_99) we utilized to address attacks on the network. Subsequently, we preprocessed the dataset by employing normalization and one-hot encoding for input into the DNN model. The refined data underwent the application of the DNN algorithm to construct a learning model, and the validation was conducted using the complete dataset. The data is split into 80% training, 20% testing and taking from the training dataset 20% for evaluation of the result. Accuracy, Detection rate, loss & Precision were calculated to confirm the detection efficacy of the LSTM and MLP at (binary and multi) models, which we found to generate performance-acceptable results for intrusion detection. The average Accuracy of the four models is 99.99% and loss is minimum. (This study was produced from the "Master's Thesis" of the first-ranked author at Yıldız Technical University, Institute of Science and Technology. ORCID ID: 0009-0002-0373-7895)
Anahtar Kelimeler: Intrusion Detection System (IDS), Deep Learning, LSTM, MLP, Binary class, Multiclass classification
|
