Network traffic prediction plays a crucial role in optimizing resource allocation, mitigating congestion, and enhancing cybersecurity in large-scale Internet of Things systems. However, the multiscale temporal patterns, complexity, and nonlinearity of traffic data pose significant challenges for accurate modeling and prediction. To address these limitations, this article proposes a novel approach that combines spectrum-aided traffic decomposition with the deep learning (DL) method for network traffic prediction. Specifically, spectrum-aided traffic decomposition is performed using the seasonal-trend decomposition using loess algorithm to decompose traffic data into interpretable seasonal, trend, and residual components, where the seasonal period is dynamically optimized through frequency domain analysis. Each component is then modeled separately using a DL architecture, which is the gated recurrent units-based sequence-to-sequence model with an attention mechanism. This allows the model to effectively capture multiscale temporal dependencies, long-term relationships, and complex patterns, thereby enhancing prediction accuracy. The predictions from each decomposed component are subsequently ensembled to obtain the final prediction value. Experimental results show that our method achieves significant improvements over other methods across four real-world datasets, reducing mean square error and mean absolute error by an average of 97.6% and 84.7%, respectively.

Abstract In practical scenarios, security events triggered by abnormal network traffic often result from the collective behavior of multiple data streams, embodying group security events with collective characteristics. Existing research methods, focusing on individual data streams, lack a macroscopic analysis and struggle with challenges of analyzing massive, imbalanced data sets. To address these challenges, this paper adopts a multi-instance learning approach, mapping multiple data streams into a bag with a coarse-grained approach, where each bag corresponds to a security event label and each data stream represents an instance. We propose a multi-instance network traffic conversion method, Bag2Image, which transforms temporal multi-instance network traffic data into image representations, preserving the spatio-temporal characteristics of instances within the bag through image channels and pixels. This strategy allows the network security event prediction task to be approached as an image classification problem, leveraging advanced image classification techniques for prediction. Our cross-experiments with six advanced multi-instance learning (MIL) algorithms and six different classification models demonstrate the superior performance of our method on both the UNSW-NB15 dataset and a private dataset. Specifically, our method achieved the highest F1 scores of 77.9% and 74.4% on these datasets, respectively, representing improvements of 4.1% and 13.5% over the second-best MIL algorithm. The recall rates also saw increases of 4.1% and 13.2%, respectively.

The development of the Internet of Vehicles (IoV) has significantly enhanced connectivity and cooperation among road entities, leading to a more efficient, economical, and safer intelligent transportation system (ITS). However, this increased connectivity also exposes vehicles to a growing risk of cybersecurity threats through intravehicle and intervehicle networks. To secure IoV networks, many studies have focused on using intrusion detection systems (IDSs) based on deep learning methods to effectively detect cyber-attacks due to their ability to learn from large-scale data. Nonetheless, most existing IDSs rely on expert knowledge to manually design features, resulting in difficulties adapting to evolving attacks and information loss. To mitigate these limitations, this article presents a tokenization representation and attention mechanism-based convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) intrusion detection method. For feature extraction, we tokenize original traffic using natural language processing technique to represent discrete hexadecimal bytes as words, thus alleviating the need for manual feature design and allowing for the direct extraction of sequence patterns. For classification, we incorporate an attention mechanism into the CNN and BiLSTM architectures to enhance the accuracy of intrusion detection by focusing on critical information and capturing sequential patterns. The effectiveness of the proposed IDS is evaluated in both intravehicle and intervehicle network scenarios. Experimental results show that our method can detect various types of attacks with 100% accuracy on the Car-Hacking data set for the intravehicle network scenario. In the intervehicle network scenario utilizing the CICIoT2023 data set, our approach also achieves a high accuracy of 98%, outperforming existing methods.

To enhance the dependability of lightweight blockchain, this paper presents a privacy protection mechanism in open networks with a secure ledger for each cell of society. We propose a swarm intelligence scheme called hierarchical RAFT (HRAFT), which accumulates valuable workloads about the credibility of all the nodes to sort for selecting candidates. Thus, the scheme achieves high-performance decentralization by progressive evolution consensus.

The edge computing node plays an important role in the evolution of the artificial intelligence-empowered Internet of things (AIoTs) that converge sensing, communication, and computing to enhance wireless ubiquitous connectivity, data acquisition, and analysis capabilities. With full connectivity, the issue of data security in the new cloud-edge-terminal network hierarchy of AIoTs comes to the fore, for which blockchain technology is considered as a potential solution. Nevertheless, existing schemes cannot be applied to the resource-constrained and heterogeneous IoTs. In this paper, we consider the blockchain design for the AIoTs and propose a novel classified ledger framework based on lightweight blockchain (CLF-LB) that separates and stores data rights at the source and enables a thorough data flow protection in the open and heterogeneous network environment of AIoT. In particular, CLF-LB divides the network into five functional layers for optimal adaptation to AIoTs applications, wherein an intelligent collaboration mechanism is also proposed to enhance the across-layer operation. Unlike traditional full-function blockchain models, our framework includes novel technical modules, such as block regenesis, iterative reinforcement of proof-of-work, and efficient chain uploading via the system-on-chip system, which are carefully designed to fit the cloud-edge-terminal hierarchy in AIoTs networks. Comprehensive experimental results are provided to validate the advantages of the proposed CLF-LB, showing its potentials to address the secrecy issues of data storage and sharing in AIoTs networks.