Applied Soft Computing最新文献

{"title":"A dynamic learning-based genetic algorithm for scheduling resource-constrained projects with alternative subgraphs","authors":"Rojin Nekoueian ,&nbsp;Tom Servranckx ,&nbsp;Mario Vanhoucke","doi":"10.1016/j.asoc.2025.113316","DOIUrl":"10.1016/j.asoc.2025.113316","url":null,"abstract":"<div><div>Genetic algorithms (GAs) are population-based algorithms widely applied for solving complex scheduling problems and such the resource-constrained project scheduling problem with alternative subgraphs (RCPSP-AS) in which alternatives for work packages should be selected prior to project scheduling. The objective of this research is twofold. First, we develop a dynamic GA based on a new hybridisation of initialisation procedures, local searches based on learning approaches and restart schemes for scheduling problems in general. Second, we improve existing benchmark solutions for a large artificial dataset for the RCPSP-AS in particular. Our dynamic GA leverages existing constructive heuristics and priority rules to create a pool of high-quality initial solutions. Subsequently, these solutions are further improved by means of learning approaches that are designed as weight- or population-based local searches. In order to avoid getting stuck in a local optimum, various restart schemes are implemented. Based on our results, gradual learning and learning based on the population outperform other approaches for high-complex problem instances. Since metaheuristics — such as GAs — are mainly beneficial in complex problem settings, we are convinced that these research findings can inspire researcher when solving similar or other scheduling problems.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"180 ","pages":"Article 113316"},"PeriodicalIF":7.2,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144212206","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI software selection for cybersecurity auditing using neutrosophic CRITIC CODAS","authors":"Fatih Sahin ,&nbsp;Akin Menekse ,&nbsp;Selman Yilmaz","doi":"10.1016/j.asoc.2025.113295","DOIUrl":"10.1016/j.asoc.2025.113295","url":null,"abstract":"<div><div>Today’s companies must develop creative solutions to counter the risks of cyberattacks that make it difficult to protect their valuable information in an increasingly complex digital world. In this context, cybersecurity audits have gained importance, and companies have become especially interested in artificial intelligence (AI)-based cybersecurity audit tools. On the other hand, the selection of AI software includes multiple criteria and alternatives, and decision experts may have uncertainty in their linguistic evaluations. In this study, a new neutrosophic CRiteria Importance Through Intercriteria Correlation (CRITIC) integrated COmbinative Distance-based ASsessment (CODAS) methodology is proposed for selecting AI software for cybersecurity auditing. The importance weights of the criteria are directly calculated with the CRITIC method, and the alternatives are ranked with the CODAS approach. The uncertainty of decision experts is modeled with neutrosophic sets through truth, indeterminacy, and falsity degrees. The study includes sensitivity analyses for criterion and decision expert weights, as well as a comparative study with rank correlation analysis. Implications and discussions, limitations, and future research avenues are also given in the study.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"179 ","pages":"Article 113295"},"PeriodicalIF":7.2,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144196194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Injecting object pose relationships into image captioning via attention capsule networks","authors":"Hong Yu,&nbsp;Yuanqiu Liu,&nbsp;Hui Li,&nbsp;Xin Han,&nbsp;Han Liu","doi":"10.1016/j.asoc.2025.113310","DOIUrl":"10.1016/j.asoc.2025.113310","url":null,"abstract":"<div><div>Image captioning is a fundamental bridge linking computer vision and natural language processing. State-of-the-art methods mainly focus on improving the learning of image features using visual-based attention mechanisms. However, they are limited by the immutable attention parameters and cannot capture spatial relationships of salient objects in an image adequately. To fill this gap, we propose an Attentive Capsule Network (ACN) for image captioning, which can well utilize the spatial information especially positional relationships delivered in an image to generate more accurate and detailed descriptions. In particular, the proposed ACN model is composed of a channel-wise bilinear attention block and an attentive capsule block. The channel-wise bilinear attention block helps to obtain the 2nd order correlations of each feature channel; while the attentive capsule block treats region-level image features as capsules to further capture the hierarchical pose relationships via transformation matrices. To our best knowledge, this is the first work to explore the image captioning task by utilizing capsule networks. Extensive experiments show that our ACN model can achieve remarkable performance, with the competitive CIDEr performance of 133.7% on the MS-COCO Karpathy test split.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"179 ","pages":"Article 113310"},"PeriodicalIF":7.2,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144196173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Metric tools for sensitivity analysis with applications to neural networks","authors":"Jaime Pizarroso,&nbsp;David Alfaya ,&nbsp;José Portela,&nbsp;Antonio Muñoz","doi":"10.1016/j.asoc.2025.113300","DOIUrl":"10.1016/j.asoc.2025.113300","url":null,"abstract":"<div><div>As Machine Learning models are considered for autonomous decisions with significant social impact, the need to understand how these models work rises rapidly. Explainable Artificial Intelligence (XAI) aims to provide interpretations for predictions made by Machine Learning models, in order to make the model trustworthy and more transparent for the user. For example, selecting relevant input variables for the problem directly impacts the model’s ability to learn and make accurate predictions. One of the main XAI techniques to obtain input variable importance is the sensitivity analysis based on partial derivatives. However, existing literature of this method provides no justification of the aggregation metrics used to retrieved information from the partial derivatives. In this paper, a theoretical framework is proposed to study sensitivities of ML models using metric techniques. From this metric interpretation, a complete family of new quantitative metrics called <span><math><mi>α</mi></math></span>-curves is extracted. These <span><math><mi>α</mi></math></span>-curves provide information with greater depth on the importance of the input variables for a machine learning model than existing XAI methods in the literature. We demonstrate the effectiveness of the <span><math><mi>α</mi></math></span>-curves using synthetic and real datasets, comparing the results against other XAI methods for variable importance and validating the analysis results with the ground truth or literature information.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"180 ","pages":"Article 113300"},"PeriodicalIF":7.2,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144223329","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-view deep support vector machines based on discriminative contrastive loss","authors":"Yanfeng Li ,&nbsp;Junqi Lu ,&nbsp;Xijiong Xie","doi":"10.1016/j.asoc.2025.113296","DOIUrl":"10.1016/j.asoc.2025.113296","url":null,"abstract":"<div><div>Contrastive learning is a rapidly evolving direction due to its ability to learn outstanding discriminative representations. However, the two theoretically complementary models, contrastive learning and support vector machine (SVM), have never been integrated. In this paper, we propose two novel multi-view deep SVMs models based on the discriminative contrastive loss to solve the multi-view multi-class classification problem. Specifically, first, we impose the discriminative contrastive loss to learn the local structural information of each view. In addition, we propose a self-learning view-weight method to explore inter-view diversity information by assigning different view weights to each view, and explore cross-view consistency information by imposing similarity constraints on the disagreements generated by different view classifiers. Finally, two novel models take the <span><math><mrow><mi>o</mi><mi>n</mi><mi>e</mi><mo>−</mo><mi>v</mi><mi>s</mi><mo>−</mo><mi>r</mi><mi>e</mi><mi>s</mi><mi>t</mi></mrow></math></span> method to solve the multi-class classification problem. In the optimization problem, the back propagation method is used to implement joint learning of the entire multi-view deep model. The experimental results validate the effectiveness of the models by comparing them with state-of-the-art algorithms.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"179 ","pages":"Article 113296"},"PeriodicalIF":7.2,"publicationDate":"2025-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144189598","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The consensus and dissent model in the graph model for conflict resolution with interval fuzzy preferences with application to doctor-patient disputes","authors":"Dayong Wang ,&nbsp;Yejun Xu","doi":"10.1016/j.asoc.2025.113392","DOIUrl":"10.1016/j.asoc.2025.113392","url":null,"abstract":"<div><div>Strategic conflicts, such as doctor–patient disputes, have become a major societal concern, with increased media attention exacerbating tensions among stakeholders. A key challenge in such conflicts is information asymmetry, which leads to uncertainty in the expression of decision-maker (DM) preferences. Existing consensus models in conflict analysis typically assume clear and consistent DM preferences, limiting their applicability in real-world scenarios characterized by ambiguity and complexity. Thus, within the framework of the graph model for conflict resolution (GMCR), this paper first attempts to characterize DM’s uncertainty and resolve conflicts by considering interval fuzzy consensus and dissent preference between two DMs. The innovation of this work lies in promoting the research of consensus and dissent model in GMCR in uncertain fields. Specifically, the first work reflects the DM’s real judgments over states using interval fuzzy scales and then converts them into clear values, which can be applied to conflict analysis process. More importantly, the logical stability definitions and matrix stability definitions of the consensus and dissent model in the GMCR framework under interval fuzzy preference relations (IFPRs) for two DMs are determined. In addition, we introduce the specific solving and analysis steps for resolving real life conflicts by using proposed models. Compared to existing consensus models in conflict decision-making with crisp preferences, DM’s IFPRs in new proposed GMCR provides a new way to characterize DM’s uncertainty and constructs a set of stability definitions in uncertainty decision making situations. Finally, in order to illustrate the correctness and scientificity of the new proposed GMCR model, it is applied to real-life doctor-patient disputes in China. The model’s validity and applicability are demonstrated through a case study of doctor–patient disputes in China, with the stability analysis offering practical insights for conflict resolution in uncertain environments.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"179 ","pages":"Article 113392"},"PeriodicalIF":7.2,"publicationDate":"2025-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144196195","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-objective closed-loop supply chain inventory model with learning and forgetting under carbon emission policies using NSGA-II, MOPSO, and TOPSIS","authors":"Tanmay Halder,&nbsp;Bijoy Krishna Debnath","doi":"10.1016/j.asoc.2025.113291","DOIUrl":"10.1016/j.asoc.2025.113291","url":null,"abstract":"<div><div>This study investigates the impact of workers’ good practices in remanufacturing, manufacturing, and inspection processes under the learning and forgetting (LaF) framework on total cost and carbon emissions in a closed-loop supply chain (CLSC) inventory model. The investigation is conducted under three carbon emission reduction policies: carbon tax (CT), cap-and-offset (CCO), and cap-and-trade (CCT). Workers’ involvement in the continuous learn-forget-learn process across different tasks in the CLSC, including remanufacturing, manufacturing, machine operation, inspection, and correcting production errors, boosts productivity and process quality. The main focus for the CLSC participants is sustainability, emphasizing the improvement of worker experience to enhance productivity and process quality, aiming to minimize total cost and carbon emissions. First, the multi-objective optimization problems are formulated under the CT, CCO, and CCT policies while incorporating LaF effects. The total cost function serves as the first objective, while the carbon emission function constitutes the second. The non-dominated sorting genetic algorithm II (NSGA-II) and multi-objective particle swarm optimization (MOPSO) are employed to solve the optimization model, with their parameters fine-tuned using Taguchi analysis. Pareto fronts are generated to identify optimal solutions, and the best solutions are selected using multi-criteria decision analysis (MCDA) with the technique for order preference by similarity to an ideal solution (TOPSIS). A statistical analysis is conducted to compare the performance of NSGA-II and MOPSO. Numerical results reveal that learning significantly reduces total costs and carbon emissions across all three policies. A comparative analysis of the policies with and without the LaF effect indicates that the CCT policy with LaF is the most effective in reducing total costs and emissions in the CLSC. Sensitivity analysis further highlights the impact of parameter variations on total costs and carbon emissions under different policies. As the learning exponent (LE) increases from 0 to 0.415, total costs and carbon emissions steadily decline. Under the CT policy, average costs decrease by 2.65%, while carbon emissions are reduced by 4.76%. The CCO policy results in reductions of 2.58% in costs and 5.86% in emissions. In contrast, the CCT policy exhibits the most significant improvements, with cost reductions of 3.84% and emission reductions of 6.93%.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"180 ","pages":"Article 113291"},"PeriodicalIF":7.2,"publicationDate":"2025-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144270868","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-layer stacked residual coordinate termite alate network for multi-class lung diseases detection from chest X-ray images","authors":"Raju Egala ,&nbsp;M.V.S. Sairam","doi":"10.1016/j.asoc.2025.113393","DOIUrl":"10.1016/j.asoc.2025.113393","url":null,"abstract":"<div><div>The World Health Organization identifies COVID-19, pneumonia, tuberculosis, and pneumothorax among the higher effects of death worldwide. Common symptoms include shortness of breath, fever, sneezing, and coughing. Traditional diagnostic methods such as thorough blood counts, the Mantoux skin test, antibodies, and DNA testing are often time-consuming and have a sensitivity of only about 80 %, with a 20 % error rate. As a faster and more reliable alternative, chest X-ray imaging is increasingly used for the detection of lung diseases. This research proposed a new system called the Multi-Layer Stacked Residual Coordinate Network, designed to accurately classify various lung diseases using chest X-ray images. Images are drawn from 6 datasets that are openly accessible. To enhance image quality, we develop the Gaussian Fourier Pyramid for Local Laplacian Filter technique, which combines adaptive histogram equalization with a multi-resolution Gaussian pyramid to better highlight important lung features while reducing noise. Next, the Mantis Search algorithm, a novel thresholding method, is used to segment critical regions in the X-rays, focusing the analysis on the most relevant areas. For deeper feature extraction, the model employed a Multi-Generative Adversarial Transformer, which captures complex patterns in the segmented regions. Finally, for classification, the Multi-Layer Stacked Residual Coordinate Network is optimized using Termite Alate Optimization, a metaheuristic inspired by termite foraging behavior that fine-tunes network parameters for better accuracy. According to experimental findings, the suggested method achieves 99 % accuracy, significantly outperforming existing techniques for multiclass lung disease classification.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"179 ","pages":"Article 113393"},"PeriodicalIF":7.2,"publicationDate":"2025-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144196174","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrated personalized decision method with q-rung orthopair fuzzy data for underground natural gas storage site decisions","authors":"Raghunathan Krishankumar ,&nbsp;Fatih Ecer ,&nbsp;Pratibha Rani ,&nbsp;Dragan Pamucar ,&nbsp;Serhat Yüksel ,&nbsp;Hasan Dinçer","doi":"10.1016/j.asoc.2025.113384","DOIUrl":"10.1016/j.asoc.2025.113384","url":null,"abstract":"<div><div>Location selection for underground natural gas storage is a multifaceted decision-making problem, as diverse factors are involved. Earlier studies on location selection for natural gas faced challenges such as uncertainty handling, methodical estimation of experts' reliability, capturing hesitation during factor significance calculation, and personalized location ordering. Therefore, the present work develops a novel integrated weighted aggregated sum product assessment (WASPAS) methodology with generalized (q-rung orthopair) fuzzy information, considering three dimensions of uncertainty: membership grade, hesitancy grade, and non-membership grade, with a flexible window allowing experts easy preference articulation. The reliability of experts is calculated using the Cronbach measure, and the importance of the criteria is computed based on the regret factor. A ranking algorithm is developed with a modified weighted aggregated sum product assessment formulation and choice vector to obtain personalized ordering of natural gas locations. The usefulness is illustrated using a case study of location selection for underground natural gas storage in India. Results show that, political acceptance is the most crucial indicator when selecting an optimal underground storage location for natural gas. The outcomes concluded that the introduced integrated framework (i) is robust, even after alterations are realized for the weights of the criteria and strategy values, (ii) produces rank orders that are consistent with the earlier models, and (iii) yields broader rank values, to support better discrimination of alternative locations and appropriate backup management compared to the extant model. Finally, the benefits, shortcomings, and implications are discussed. The model introduced can be a novel guide for natural gas location selection and can aid investors in planning their investments.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"180 ","pages":"Article 113384"},"PeriodicalIF":7.2,"publicationDate":"2025-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144223317","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Intelligent optimization of e-commerce order packing using deep reinforcement learning with heuristic strategies","authors":"Kaibo Liang ,&nbsp;Man Shan ,&nbsp;Huwei Liu ,&nbsp;Jianglong Yang ,&nbsp;Chenxi Gu ,&nbsp;Xiangyu Yin","doi":"10.1016/j.asoc.2025.113283","DOIUrl":"10.1016/j.asoc.2025.113283","url":null,"abstract":"<div><div>The rapid expansion of e-commerce has intensified demands for efficient logistics, particularly in optimizing three-dimensional bin packing (3D-BPP) to balance space utilization, operational costs, and sustainability. Traditional methods often fail to address the dynamic, multi-constrained nature of e-commerce orders, which involve diverse item combinations, real-time decision-making, and complex practical constraints. This study proposes a hybrid framework that integrates deep reinforcement learning (DRL) with heuristic strategies to tackle these challenges. We first formulate a comprehensive mathematical model for 3D-BPP that explicitly incorporates rotation, boundary, and non-overlapping constraints. Building on this foundation, we develop a heuristic strategy system with five operator components for bin selection, item grouping, packing sequence, position selection, and orientation determination. To enhance adaptability, we introduce two DRL algorithms: the Order Packing Optimization DRL (OPO-DRL) for dynamic item sequencing and the Packing Combination Strategy DRL (PCS-DRL) for adaptive operator selection. The hybrid framework synergizes DRL’s learning capabilities with heuristic efficiency, enabling real-time adjustments to varying order patterns and bin specifications. Experimental validation using real-world data from JD.com demonstrates significant improvements, achieving an average packing rate of 68.60% with computation times of 0.16 s per order, outperforming state-of-the-art methods. Statistical analysis confirms significant improvements in both solution quality and computational efficiency compared to existing approaches. This work bridges theoretical optimization with operational realities, providing a scalable solution for modern warehouse automation and intelligent logistics systems.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"179 ","pages":"Article 113283"},"PeriodicalIF":7.2,"publicationDate":"2025-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144204808","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}