Electrical Engineering Technical Journal

The rapid expansion of Industrial Internet of Things (IIoT) networks has significantly increased the vulnerability of industrial systems to diverse cyber-attacks. Therefore, designing an intelligent and adaptive intrusion detection model capable of handling high-dimensional, streaming, and dynamically evolving network data is essential. In this paper, we propose a novel intrusion detection framework based on the Adaptive Streaming Decision Tree (ASDT) algorithm integrated with Neighborhood Component Analysis (NCA) for optimal feature selection. First, a preprocessing stage is applied to the NSL-KDD and UNSW-NB15 benchmark datasets, including data normalization and outlier removal, to enhance data consistency and reduce noise. Then, NCA-based feature selection is employed to identify the most discriminative attributes, effectively reducing computational complexity and improving classification performance. Finally, the selected features are fed into the proposed ASDT classifier, which incrementally learns from network traffic streams and dynamically adapts to concept drift through a forgetting-factor mechanism and online entropy-based splitting criteria. Experimental results demonstrate that the proposed method achieves outstanding detection performance, with an average accuracy of 99.34% on the NSL-KDD dataset and 99.02% on the UNSW-NB15 dataset, outperforming conventional decision tree and ensemble-based intrusion detection models. The results confirm that the proposed ASDT-NCA framework provides a robust, interpretable, and adaptive solution for real-time IIoT network intrusion detection.

Subclinical seizures are subtle neurologic events that remain largely undetectable during standard electroencephalography (EEG), and thus present critical barriers to timely diagnosis and monitoring of affected individuals. In this paper, we propose a deep learning framework for automatic detection of subtle patterns in EEG signals that are indicative of subclinical seizure activity. The available high-quality EEG data were preprocessed with band-pass and notch filters, ICA-based artifact removal, as well as splitting into fixed-sized epochs. Multiple input representations including raw signals, spectrograms, and wavelet transforms were used to broaden the feature extraction power of the model. A variety of deep learning architectures (such as CNNs, LSTM networks and hybrid CNN-LSTM models) were all designed and trained based on supervised training. The performance of the models was assessed using accuracy, sensitivity, specificity, F1-score and AUC. The hybrid CNN–LSTM model outperformed, with high sensitivity and accuracy in the detection of sub-clinical seizures across subjects. The obtained results suggest that deep learning can effectively encode complex spatiotemporal patterns observed in EEG activity although they are often overlooked by traditional methods, serving as a reliable, unbiased tool for neurological diagnostics. This work demonstrates that it may be feasible for intelligent EEG analysis systems to aid in 24/7 patient monitoring, improve clinical management decisions and interventions, not only when seizures are full-blown but also in the early intervention and treatment of subclinical seizure control.

The fifth generation (5G) network represents an advancement in mobile communication technology in a region with increasing demands in usage to solve the problem of traffic congestion, high latency, and low speed. However, interference is one of the major limiting factors that degrades the 5G network performance. The adverse effects of interference have led to several research studies into mitigation techniques for 5G networks. This paper presents a comprehensive review of interference types and mitigation techniques in 5G networks, including conventional approaches such as power control, inter-cell interference coordination, beamforming, and resource allocation. Moreover, this work provides an analytical framework for Deep Learning (DL)-Based Channel Estimation and Successive Interference Cancellation (SIC) in 5G non-orthogonal multiple access (NOMA) systems. This study extends existing models by incorporating residual interference modeling, highlighting the impact of error propagation in SIC systems. This paper synthesizes existing techniques, identifies key research gaps such as: imperfect channel estimation as in the conventional SIC, failure to address the error components caused by the residual interference that is often adhered to SIC technique, and establishes a unified perspective on integrating DL into interference mitigation, showing that DL-Based SIC has strong potential for improving interference resilience. We believe that our proposed technique will serve as guidelines for moving forward with SIC aware protocol research in 5G and Beyond 5G networks