Detection of denial-of-service attack using a novel hybrid learning technique

Abstract

Physical Layer Security (PLS) in wireless networks is becoming crucial with advancements in technologies like Beyond-5G (B5G) and 6G. To address growing threats such as Denial of Service (DoS) attacks, PLS uses Machine Learning (ML) techniques to detect and counter these threats effectively. PLS secures wireless communication systems, by utilizing the physical properties of the communication medium such as signal metrics, channel characteristics and noise patterns. In this paper, a novel approach to classify attack and non-attack scenarios using Long Short-Term Memory-Fully Connected network (LSTM-FCNet) for feature extraction and Gradient Boost (GB) algorithm for classification has been proposed. The DoS attack datasets are generated in the form of jamming, where both attack and non-attack case wireless channel behaviour are captured using Channel State Information (CSI) under various Signal to Noise Ratio (SNR) conditions. The proposed hybrid learning technique plays a crucial role to extract features, in order to capture temporal dependencies in the data, which is significant for identifying delicate patterns. These features are then classified using the GB algorithm to accurately distinguish between attack and non-attack scenarios. The simulated results show that the attack detection accuracy has been achieved up to a maximum of 98.25 % for different SNR values, with precision, recall, and F1-score of all achieving 98 %. The Receiver Operating Characteristic (ROC) curve with a value of 0.99 indicates that the classifier has achieved a high True Positive Rate (TPR). The results ensure that the classifier works at peak accuracy for the developed attack detection model, effectively handling the generated DoS attack dataset.