ABSTRACT Predicting concrete compressive strength (CS) and fire resistance (FR) is a crucial difficulty due to the high cost and time required for laboratory tests. This study introduces a hybrid machine learning system aimed at predicting these properties, utilizing nine models derived from 408 mix designs. The methodology utilized involved data cleaning, extensive feature engineering, k-fold cross-validation, and a separate test partition. An important advancement is the development of a dual-property prediction system that utilizes Explainable AI. The Stacking Ensemble and XGBoost algorithms exhibited remarkable accuracy, achieving R2 scores of 0.984 for CS and 0.985 for FR. The mean error recorded was 2.01 MPa for CS and 7.26 minutes for FR. The SHAP analysis clarified the results, identifying the water–cement ratio and cement content as the primary influential factors, consistent with principles of material science. A graphical user interface for a web application was developed to implement this predictive architecture for optimizing mixes in real time. This tool offers engineers prompt predictions, potentially decreasing design cycle times by 60–80% and minimizing dependence on physical testing, thereby promoting a data-driven approach in civil engineering.

ABSTRACT Current interconnected global economy, the dynamic evolution and complex interdependencies of supply chain networks create high demands for risk prediction and management. Existing methods often struggle with modeling heterogeneous structures, identifying potential risk nodes, and perceiving anomalies in dynamic scenarios. To address these issues, we propose a Relation-aware Fusion Risk Perception Network (RFRP-Net), which integrates graph neural networks with anomaly-aware attention for fine-grained risk modeling and detection. RFRP-Net introduces a Relation-Aware GCN (RA-GCN) module that constructs multi-granularity relation mechanisms based on node business similarity and risk intensity, enabling dynamic aggregation across heterogeneous nodes. To enhance anomaly perception, we design an Attention-based Abnormality Detection Feature Fusion (AADFF) mechanism, which employs anomaly query vectors with sparsity and deviation constraints. By leveraging Bi-LSTM and cross-attention, AADFF captures abnormal interactions, extracts risk features, and generates multidimensional anomaly indicators such as attention entropy and focus deviation. A final feature decoder predicts both risk factors and specific risk items. We hypothesize that RA-GCN improves the accuracy of heterogeneous risk propagation modeling, while AADFF strengthens anomaly identification by capturing weak risk signals. Experiments on real-world datasets confirm that RFRP-Net significantly outperforms mainstream methods, providing practical support for supply chain risk management. Our code is available at https://github.com/FudWang/Relation-aware-Fusion-Risk-Perception-Network.git.

ABSTRACT This paper proposes a generalized-metric-based pattern recognition method using Ordered Pairs of Normalized Real Numbers (OPNs). The proposed approach overcomes the limitations of real-valued metrics by effectively extracting high-dimensional information from feature interactions, which traditional distances cannot capture. Specifically, it employs a feature pairing that combines original features into real-number pairs to capture inherent feature correlations. A generalized metric built on OPNs theory allows effective comparison of distances expressed as real-number pairs. To address the combinatorial explosion problem in high-dimensional data, a correlation-based pre-screening strategy for pairing is introduced, significantly reducing the search space while maintaining classification performance. Experiments on multiple datasets show that this method achieves significantly higher accuracy than traditional distances and some deep learning models, reaching 95.33% on Iris and 90.00% on Seeds, with statistical tests confirming its effectiveness and stability.

ABSTRACT The process of inferring the intention of conversational implicatures involves the interpretation of the speaker’s sentiment. However, the relationship between implicatures and sentiments has not been clear enough, and there is no research explaining the relationship between the response orientation of conversational implicature and sentiment score. Therefore, based on the dialogue snippets of conversational implicature, this paper uses automatic sentiment analysis, statistical testing and other methods to identify the statistical dependencies between implicatures and sentiments, and compare the sentiment scores of implicatures and literal meanings. The results show that the response orientation of implicature has a significant impact on the sentiment score, and its source only contains response utterance. In addition, within the response utterance, whether there is a significant difference between the sentiment scores of implicature and literal meaning is related to the selected algorithms of sentiment analysis. The sentiment lexicon-based method like Pattern cannot distinguish the sentimental difference between implicature and literal meaning, but the sentiment score obtained by the VADER-based method that considers grammatical and syntactical heuristics has significant differences in implicatures and literal meanings. Finally, the methodological implications of the experiments and results for the development of computational pragmatics are provided in this paper.

ABSTRACT Vehicular Communication Networks (VCNs) are essential for autonomous vehicles and Intelligent Transportation Systems but face challenges in security vulnerabilities and data sparsity. Traditional attack models inadequately represent VCN dynamics, weakening threat detection, while existing datasets lack real-world mobility and spatiotemporal details. This study addresses these gaps by developing a comprehensive attack simulation framework, enhancing critical network attacks i.e. position spoofing, Sybil, and wormhole through realistic mobility patterns, positional dynamics, and temporal interactions. The resulting dataset contains legitimate and malicious instances: Spoofing (45,975 legitimate, 589 malicious), Wormhole (52,237 legitimate, 5,219 malicious), and Sybil (14,829 legitimate, 1,753 malicious). It includes essential vehicular-specific features such as mobility dynamics, inter-vehicle distances, and end-to-end communication patterns. For validation, machine learning algorithms, including Random Forest, K-Nearest Neighbors, and Logistic Regression were employed. Detection performance was evaluated using accuracy, precision, recall, and two F1-score variants (standard and macro). Results indicate high detection efficacy, with Random Forest achieving accuracy between 93.6% and 99.8% and F1-macro scores from 88.5% to 97.7%. Compared to previous studies lacking spatiotemporal considerations, our dataset’s enhanced realism demonstrates significant potential in advancing data-driven anomaly detection and real-world threat mitigation in dynamic vehicular environments.

ABSTRACT Understanding and interpreting machine learning models is crucial in high-stakes industries like steel manufacturing, where decisions impact energy efficiency and environmental sustainability. Traditional feature importance methods often struggle with robustness under noisy conditions, leading to unreliable insights. To address this problem, we introduce the Complexity and Interaction Enhanced SHAP (CESHAP), a novel feature importance method that incorporates model complexity and feature interactions. Alongside, we propose the Noise-Adjusted Feature Importance Change (NAFIC) metric to assess the robustness of feature importance methods against varying levels of noise. Experiments conducted on an energy consumption dataset from the steel industry, with systematically introduced Gaussian noise levels (5%, 10%, 15%, 20%), demonstrate that CESHAP offers superior robustness in tree-based models compared to SHapley Additive exPlanations (SHAP) and Permutation Feature Importance (PFI). Our findings underscore the effectiveness of CESHAP in enhancing interpretability and reliability in complex, non-linear models, ultimately supporting more informed decision-making in energy-intensive industries.

ABSTRACT With the rapid development of autonomous vehicles and mobile robotics, lidar has been one of the most popular researches in the world. But the poor ranging accuracy and detection range in the foggy situation have limited the application of lidar. In this paper, the detection of pulsed lidar in fog is investigated. The transmission model in the fog of the laser signal is established by considering laser backscattering in the fog. A Prior Waveform Guided Network (PWGN) based on the Convolutional Neural Network (CNN) is proposed. The results of simulation and experiments show that the PWGN can effectively remove the interference of fog in the detection of pulsed laser signal. The mean of absolute errors(MAE) of the detection achieves 3.13 cm at the range of 10 m at the scattering rate of 30%. The MAE may be half or 1/3 of the MAEs of other approaches at the detection range of 42 m.

ABSTRACT Remote diagnosis enables healthcare professionals to evaluate and diagnose patients from a distance using telecommunication technologies, enhancing healthcare delivery by improving accessibility, especially for those in remote or underserved areas. One of the significant sustainability challenges in remote medical diagnostics is offering timely assistance to vulnerable groups like the elderly, disabled, mentally impaired individuals, and wounded military personnel in combat zones. This becomes particularly difficult in emergencies when rapid analysis of medical records is needed, especially if the data is stored on secure blockchain networks. The proposed work addresses these challenges by deploying a comprehensive framework for large-scale analysis, utilizing both document and image classification for dual validation. It integrates advanced techniques such as Inception V3, VGG-16, VGG-19, RESNET-50, and Densenet-201 for bone fracture detection, with Inception V3 achieving the highest accuracy of 95.1%. In addition, a Document Classification Analysis (DCA) method is proposed, which automatically classifies the severity of fractures. Object detection techniques are also introduced for detecting minor fractures using region-based image segmentation, ensuring precise diagnosis even for subtle injuries. This pioneering integration of technologies provides a holistic solution for remote medical diagnostics.

ABSTRACT One of the issues addressed by machine learning, with applications in various disciplines or fields such as the health sector and the agricultural sector among others, involves data classification. For this purpose, various models within supervised learning have been proposed and developed that allow for the classification of these data. However, one of the implications of the No-Free-Lunch theorems is that there is no optimal general-purpose model, i.e. there is no classifier model that achieves the best results for all problems presented. Hence the importance of proposing and implementing new classifier models, evaluating their performance, and comparing them with other classifier models in order to achieve models with good results in specific problems. This work presents a new classifier model that, by constructing hyperplanes from the training set, generates a decision tree and partitions the dimensional space. The proposed model was applied to different problems such as the XOR logical function problem, where the proposed model managed to solve the problem, it was also applied to the Iris Dataset where one of the trees generated by the model managed to classify with 100% accuracy the test set, and finally, the proposed model was applied to the Pima Indians Diabetes Database and compared with other models using the accuracy value. The proposed model obtained an accuracy of 81.81%, achieving the best result in the same way as the Random Forest Classifier. The results obtained in this work show that the proposed model manages to partition the dimensional space adequately from the training set and thus competitively classify the data with other state-of-the-art models.

ABSTRACT This research focuses on addressing the challenges surrounding interpretability of machine learning techniques in the field of prediction of disease progression. This paper summarizes the state-of-the-art in machine learning for disease progression modeling and challenges related to this context. Based on this state-of-the-art, we design and develop a pipeline consisting of data preparation, prediction, and explanation. Predictions are made using a deep recurrent neural network-based model which is followed by an integration of the LIME framework to provide explanations for each prediction. We demonstrate our pipeline by applying it to two diverse case studies on diabetes and Parkinson’s disease. Besides this, we compare the influence of three data imputation methods on predictive performance. Results of the comparison show that there is no statistically significant difference in performance due to different data imputation techniques. Furthermore, we provide a number of concrete recommendations and directions for future research, such as improving input flexibility of the prediction model and improving the visualization of generated explanations. Based on the results of this research, we conclude that the proposed pipeline achieves the goal of integrating a state-of-the-art prediction model and the LIME framework to make prediction of disease progression more transparent and interpretable.