Internet of things (IoT) networks increasingly need security due to the large amount of data that needs to be managed. These networks are susceptible to a variety of sophisticated and more frequent cyberattacks. In th...
详细信息
Internet of things (IoT) networks increasingly need security due to the large amount of data that needs to be managed. These networks are susceptible to a variety of sophisticated and more frequent cyberattacks. In this study, an improved cyber-attack detection model is presented for IoT networks using a fine-tuned deep learning model. This model produces high accuracy and classifies the different types of cyber-attacks with low losses. In the feature selection process, a wrapper-based dwarf mongoose optimisation algorithm (W-DMO) is utilised to choose the best subset of features from the original network traffic features. Lastly, a hybrid triple attention deep neural network-assisted BiLSTM model (TDeepBiL) is employed to classify the features and categorise different kinds of attacks. Several performance metrics are evaluated for the proposed method, including accuracy, precision, recall, and F1-score. The proposed model has reached a high accuracy of 99.44% for the UNSW-NB 15 dataset and 98.6% for the ToN-IoT dataset in comparison to other current models. Thus, the presented model gains significant improvement in cyber-attack detection.
This research paper focuses on the domain of Acoustic Anomaly Detection (AAD) in industrial machinery using Deep Learning techniques. The primary objective of this study is to develop a reliable system for detecting a...
详细信息
ISBN:
(纸本)9798350371345
This research paper focuses on the domain of Acoustic Anomaly Detection (AAD) in industrial machinery using Deep Learning techniques. The primary objective of this study is to develop a reliable system for detecting anomalies in the acoustic patterns of industrial machines. In large-scale industrial environments, the early detection of faults and anomalies is critical to ensure the uninterrupted operation of machinery, minimize downtime, and optimize maintenance efforts. The datasets used in this research encompass a range of Signal-to-Noise Ratios (SNR) to simulate diverse operating conditions for industrial fans and pumps. After preprocessing each audio sample was transformed into 9 segments of shape 128 x 32 log mel-spectrograms. The study encompasses a comprehensive analysis of Convolutional autoencoders (CAE) and dense autoencoders (DAE). These models are trained and evaluated against real-world industrial datasets (MIMII dataset), and their performance is meticulously compared. The results for the DAE and CAE had shown over 0.80 at the 6 dB SNR level and decaying results as the SNR level worsens. This research confirms some of the trends pointed out by the literature and provides detailed insight into how the autoencoders are developed and their properties could be used in order to detect anomalous behavior in audio data.
暂无评论