The proliferation of IoT devices has expanded the attack surface, making zero-day threat detection an urgent priority. This paper proposes LLMEdgeSec, a framework that leverages large language models (LLMs) for log reasoning to detect and contextualize zero-day exploits in real-time. Unlike conventional anomaly detectors that rely on static signatures or pre-defined rules, LLMEdgeSec applies prompt-based reasoning over system logs, communication metadata, and device telemetry, enabling proactive identification of previously unseen attack vectors. The framework employs a dual-stream architecture: (1) a lightweight transformer encoder on edge nodes for local feature extraction, and (2) a cloud-based LLM interface for contextual reasoning and cross-device threat correlation. A zero-shot calibration method mitigates hallucinations and reduces false positives. Evaluation on TON IoT, CIC-IDS2018, and a real-world smart factory dataset demonstrates F1 scores exceeding 92% on zero-day samples with communication overhead below 8MB per federated round. Results show significant improvements in accuracy, interpretability, and adaptability over state-of-the-art baselines.

Driver drowsiness poses a significant threat to road safety, contributing to numerous accidents globally, according to historical statistical data. This study provides an exhaustive overview of driver drowsiness, encompassing its symptoms, causes, prevention strategies, and underlying physiological and neural changes that occur when transitioning from wakefulness to a drowsy state. This review paper explores the complexities of detecting driver drowsiness, with a focus on brain imaging-based methodologies, and addresses four key research questions. We systematically analyze and review existing research on driver drowsiness detection using machine learning algorithms at both macro and micro levels, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. We categorize brain imaging modalities into four main groups: electroencephalography (EEG), functional near-infrared spectroscopy (fNIRS), functional magnetic resonance imaging, and magnetoencephalography. Micro analysis (from 2020 to 2025) explores the application of these modalities in detecting driver drowsiness, discussing their strengths and limitations, and evaluating their effectiveness in simulated and real-world driving experiments. Our evaluation reveals that EEG and fNIRS have emerged as the most prevalent brain imaging methodologies for detecting driver drowsiness, due to their non-invasive and portable nature, high temporal and spatial resolution, and real-time capability. However, challenges persist, including the need for more robust machine learning algorithms, improved signal processing techniques, and enhanced sensor technologies. Furthermore, there is a need for more comprehensive studies that integrate multiple brain imaging modalities and machine learning approaches to detect driver drowsiness. By addressing these key areas, future research can advance the field of driver drowsiness detection and pave the way for safer and more efficient transportation systems.

Quantum information processing introduces novel approaches for classical data encoding to encompass the complex patterns of input data of practical computational challenges using basic principles of quantum mechanics. The classification of diabetes is an example of a problem that can be efficiently resolved by using quantum unitary operations and the variational quantum classifier (VQC). This study demonstrates the effects of the number of qubits, types of feature maps, optimizers’ class, and the number of layers in the parametrized circuit, and the number of learnable parameters in ansatz influences the effectiveness of the VQC. In total, 76 variants of VQC are analyzed for four and eight qubits’ cases and their results are compared with six classical machine learning models to predict diabetes. Three different types of feature maps (Pauli, Z, and ZZ) are implemented during analysis in addition to three different optimizers (COBYLA, SPSA and SLSQP). Experiments are performed using the PIMA Indian Diabetes Dataset (PIDD). The results conclude that VQC with six layers embedded with an error correction scaling factor of 0.01 and having ZZ feature map and COBYLA optimizer outperforms other quantum variants. The optimal proposed model attained the accuracy of 0.85 and 0.80 for eight and four qubits’ cases, respectively. In addition, the final quantum model among 76 variants was compared with six classical machine learning models. The results suggest that the proposed VQC model has outperformed four classical models including SVM, random forest (RF), decision tree (DT), and linear regression (LR).

The global metal detector (MD) market is predicted to grow to U.S.${\$}$2.74 billion by 2030, with a compound annual growth rate (CAGR) of 7.66%, indicating potential expansion in the industry. Metallic and non-metallic material/object detection is pivotal to various technical applications and paramount in ensuring structural integrity, operational efficiency, and overall safety. This study analyzes a multi-sensor fusion approach for classification of metallic and non-metallic objects in complex soil environments at various depths. The adopted approach is beneficial in scenarios where single-sensor systems fail to provide adequate performance. We investigated the integration of magnetic induction spectroscopy (MIS) and ground penetrating radar (GPR) sensors for binary, soil, depth, and multiclass classification. A dataset comprising 14 objects was trained and tested on six AI models (900 variants). The best-performing 384 model variants, which have the highest accuracy, precision, and recall, reaching 98%+ and with minimal Type I and II errors, are reported here. The best accuracy for each classification problem (binary, soil type, depth, and multiobject) corresponding to each set of input data (MIS, GPR, and MIS+GPR) has been summarized and reported. The analysis shows that different AI models have performed better in different scenarios. The study’s implications extend to defense, archeology, and geology, offering enhanced object detection in complex soil environments. These findings underscore the multi-sensor fusion approach’s significance for advancements in remote sensing and classification technologies.

The proposed study is based on a feature and channel selection strategy that uses correlation filters for brain-computer interface (BCI) applications using electroencephalography (EEG)-functional near-infrared spectroscopy (fNIRS) brain imaging modalities. The proposed approach fuses the complementary information of the two modalities to train the classifier. The channels most closely correlated with brain activity are extracted using a correlation-based connectivity matrix for fNIRS and EEG separately. Furthermore, the training vector is formed through the identification and fusion of the statistical features of both modalities (i.e., slope, skewness, maximum, skewness, mean, and kurtosis). The constructed fused feature vector is passed through various filters (including ReliefF, minimum redundancy maximum relevance, chi-square test, analysis of variance, and Kruskal-Wallis filters) to remove redundant information before training. Traditional classifiers such as neural networks, support-vector machines, linear discriminant analysis, and ensembles were used for the purpose of training and testing. A publicly available dataset with motor imagery information was used for validation of the proposed approach. Our findings indicate that the proposed correlation-filter-based channel and feature selection framework significantly enhances the classification accuracy of hybrid EEG-fNIRS. The ReliefF-based filter outperformed other filters with the ensemble classifier with a high accuracy of 94.77 ± 4.26%. The statistical analysis also validated the significance (p < 0.01) of the results. A comparison of the proposed framework with the prior findings was also presented. Our results show that the proposed approach can be used in future EEG-fNIRS-based hybrid BCI applications.

In 2019, a deadly coronaviral infection (COVID‐19) that infected millions of people globally was detected in China. This fatal virus affects the respiratory system and currently spreads to more than 200 nations worldwide. COVID‐19 may be found using a chest X‐ray scan, a reliable imaging method. Although an expert may examine an X‐ray scan manually, this process takes a lot of time. Therefore, deep convolutional neural networks (CNNs) may be utilized to automate this procedure. In this work, at the first step, a novel isolated 19‐layer CNN model is developed from scratch to detect chest infections using X‐rays. Then, the developed model is reutilized to distinguish the type of chest infection, such as COVID‐19, fibrosis, pneumonia, and tuberculosis, using the transfer learning approach. Stochastic gradient descent with momentum is utilized to optimize the model. The proposed multistage framework shows 98.85% and 97% classification accuracies for chest infection detection (binary classification between normal and patient) and four‐class subclassification (COVID‐19, fibrosis, pneumonia, and tuberculosis) for an online chest X‐ray dataset. The reliability of the proposed multistage CNN model was further validated through a new dataset, showing an accuracy of 98.5%. The proposed multistage methodology took minimal training time compared to publically available pretrained models. Therefore, the presented multistage deep learning framework can help doctors in clinical practices.

In this study, a channel and feature selection methodology is devised for brain‐computer interface (BCI) applications using functional near‐infrared spectroscopy (fNIRS). A graph convolutional network (GCN) is employed to select the appropriate and correlated fNIRS channels. Furthermore, in the feature extraction phase, the performance of two filter‐based feature selection algorithms, (i) the minimum redundancy maximum relevance (mRMR) and (ii) ReliefF, is investigated. The five most commonly used temporal statistical features (i.e., mean, slope, maximum, skewness, and kurtosis) are used, whereas the conventional support vector machine (SVM) is utilized as a classifier for training and testing. The proposed methodology is validated using an available online dataset of motor imagery (left‐ and right‐hand), mental arithmetic, and baseline tasks. First, the efficacy of the proposed methodology is shown for two‐class BCI applications (i.e., left‐ vs. right‐hand motor imagery and mental arithmetic vs. baseline). Second, the proposed framework is applied to four‐class BCI applications (i.e., left‐ vs. right‐hand motor imagery vs. mental arithmetic vs. baseline). The results show that the number of appropriate channels and features was significantly reduced, resulting in a significant increase in classification accuracy for both two‐class and four‐class BCI applications, respectively. Furthermore, both mRMR (i.e., 87.8% for motor imagery, 87.1% for mental arithmetic, and 78.7% for four‐class) and ReliefF (i.e., 90.7% for motor imagery, 93.7% for mental arithmetic, and 81.6% for four‐class) yielded high average classification accuracy (p < 0.05). However, the results of the ReliefF algorithm are more stable and significant.