Fostering the inclusion of vulnerable members is particularly important in team-based workplace environments, where collaboration and collective engagement are crucial for success. However, vulnerable team members may face unique challenges, necessitating a structured approach to identifying and providing them with the appropriate support. This study seeks to identify team members who may require additional support and to determine the most suitable individuals within the team to act as supporters, thereby optimizing overall team contributions. We aim to foster enhanced collaboration and ensure more effective teamwork by evaluating each member's performance in detail. Employing Interval Data Envelopment Analysis (Interval DEA), we establish efficiency intervals encompassing a range of possible efficiency values from multiple perspectives. This methodology ensures that all team members are actively contributing. This approach aligns directly with Goal 8 of the Sustainable Development Goals (SDGs), which emphasizes inclusive economic growth and the promotion of decent work for all. By fostering inclusive participation within teams, we contribute to the broader objectives of sustainable development, ensuring that every team member can reach their full potential and that team performance is maximized through collective effort. This methodology enhances individual contributions and strengthens overall team cohesion and performance.
The Neuroticism Extraversion Openness - Five Factor Inventory is extensively utilized in psychological research to quantitatively evaluate personality traits. However, psychological biases may influence test takers to intentionally fabricate responses to present themselves more favorably. This manipulation can compromise the reliability of the questionnaire results. To address this issue, we hypothesized that classifying responses as either honest or fabricated based on the influence of psychological bias could enhance the reliability of the assessment. In this study, electroencephalography data recorded during the response process was used to define information transfer in brain activity while responding by Granger causality, which was used as a feature for a support vector machine to perform a binary classification of responses as honest or fabricated. A validation experiment was conducted with 23 participants under two conditions. In one condition, participants were instructed to manipulate their responses to make them appear more favorable. In the other condition, they were asked to respond honestly. Based on the difference between the two conditions, responses were labeled according to the presence or absence of psychological bias. To evaluate the classification accuracy, we applied leave-one-subject cross-validation and evaluated the generalization performance of the model. The results showed that the accuracy was 0.797, the positive predictive value was 0.853, the negative predictive value was 0.713, the recall was 0.840, the specificity was 0.709, and the F1-score was 0.835.
Bayesian neural networks (BNNs) are competitive multiclass classifiers with reasonably sound probabilistic characteristics. In this paper, we consider BNNs as classifier ensembles obtained through sampling, and define the inference problem in BNNs as 1. computing a median probability distribution from the ensemble of distributions, under a given distance, and 2. finding a Bayes-optimal prediction (BOP) under the evaluation metric given the median. With this (re)formulation, all the results related to computing medians of sets of probability distributions can be leveraged to strengthen the predictive performance of BNNs. We shall recall a generic formulation of the problem of computing medians and provide empirical evidence to illustrate the potential impact of the choice of distance regarding accuracy metrics and calibration errors.
The article addresses the problem of sorting objects based on incomplete pairwise comparison matrices. An example of using intervalvalued fuzzy logic is the ranking of the best female tennis players according to the win/loss ratios. An incomplete pairwise comparison matrix (PCM) is constructed using information obtained from the official website of theWomen's Tennis Association (WTA). The top 29 players on the official WTA rankings have their live results from 1973 to 2024 included in the database. Using Bayesian statistics, an incomplete interval-valued fuzzy preference relation is derived from the incomplete PCM, and various optimization models are used to determine the consistent intervalvalued weight vectors.
This paper presents an enhanced iris verification system that improves segmentation and template matching by integrating additional evaluation metrics and registration screening. Traditional methods primarily use Hamming distance for template comparison, which can result in inadequate accuracy when distinguishing similar and dissimilar templates. To enhance performance, we incorporate Hamming distance, Jaccard distance, and Pearson correlation for a more comprehensive analysis, along with a variance-based enrollment screening mechanism to reject poorly segmented images. Evaluated on the CASIA-IrisV2 dataset, containing 1,200 images and 719,400 unique pairs, our results show significant improvements in balanced accuracy and recall, reducing false negatives. Although precision slightly declines, leading to increased false positives and affecting the F0.5 score, further refinement using variance thresholds mitigates this issue without compromising recall. Overall, our approach effectively addresses class imbalances and minimizes misclassification risks, resulting in a more reliable verification process capable of accommodating a broader range of input qualities while sustaining high overall performance.
Visual Question Answering (VQA) in medical imaging has the potential to transform clinical practice by enabling automated, image-based responses to medical queries, particularly in radiology. However, most Medical VQA models are large-scale and computationally demanding, making them impractical for widespread use in resource-limited healthcare settings. In this work, we introduce MiniMedMind, a lightweight and efficient VQA model designed to perform medical question-answering tasks on chest X-Ray (CXR) images. MiniMedMind combines a vision encoder, a projector layer, and a fine-tuned small-scale language model to achieve robust performance with minimal computational requirements. The vision encoder leverages pretrained weights from CheXAgent for effective feature extraction, while the projector aligns these features with a finetuned Llama 3.2 model (3B parameters) enhanced with Low-Rank Adaptation (LoRA) to generate clinically accurate responses. The training dataset combines patient-doctor conversations, MIMIC-CXR data, and synthetic conversations generated by GPT-3.5, enabling MiniMedMind to effectively interpret and respond to medical queries. Evaluated on report generation and VQA tasks, MiniMedMind performs near top models like XrayGPT and Med-MoE, achieving competitive accuracy with a significantly lighter architecture. These results position MiniMedMind as an efficient, effective solution for resource-constrained medical AI applications. MiniMedMind offers a computationally efficient solution for Medical VQA, enabling AI-driven diagnostic support that can be applied in diverse clinical and educational settings.
In semi-automated production systems, maintaining consistent productivity is challenging due to uncertainties arising from machinery and human labour losses. This research integrates Overall Equipment Effectiveness (OEE) and Overall Labour Effectiveness (OLE) to provide a comprehensive guideline for improving productivity. By applying the principles of Total Productive Maintenance (TPM), the study aims to understand and reduce system losses. Key metrics from OEE and OLE are analysed to identify losses, with particular focus on the interactions between machine and manpower losses. The approach is validated using a prototype production line, demonstrating methods to effectively collect and analyse data on various types of losses. This proposed procedures offers actionable insights for addressing problems and implementing targeted improvement activities to enhance overall productivity.
During the last hit of Coronavirus disease (COVID-19), all the countries were troubled by as they were not ready for, thus many serious consequences had popped up to the social, economic, and health aspects of our lives. Hence, people should be vaccinated as soon as possible to eradicate the pandemic. The countries faced difficulties those, generally, revolve around tracking, queue management, privacy, data security, accessibility, and other operational challenges. This project is concerned about solving the above problems by proposing a mobile application for nationwide vaccine distribution management. Three similar existing systems, namely MySejahtera, COVID-19 UAE, and Halton Region were reviewed and analyzed based on their features and weaknesses to get the requirements of the proposed solution. The important features included in the proposed solution are register and manage vaccination appointments, book vaccination for dependents, risk assessment tool, pandemic statistics, and announcements. Apart from that, the evolutionary prototyping model was used in this project to develop the proposed solution. By the deployment of such an application, the world will be ready to face any future pandemic without being troubled.
Despite a rich literature on explainable classification, to our knowledge, there is a lack of classification methods that come with reliable user-orientated explanations supporting the predictions. To complement the existing literature on explainable classification, we propose a Self-Explaining Neural Network for Multi-Criteria Sentiment Analysis (SENN4MCSA) which consists of three key components: topic modeling, which extracts relevant topics from the training data, topics-criteria alignment, which partitions the relevant topics into the evaluation criteria given by the end-user, and self-explanation sentiment analysis, which consists of training a self-explanation classification and explanation step based on domain knowledge extracted from the topics-criteria alignment phase. More precisely, the output of the topics-criteria alignment is taken into account in the explanation step to provide user-orientated explanations supporting the prediction of the self-explanation classification. We implement SENN4MCSA by employing BERTopic, manual topicscriteria alignment based on domain experience, and SELFEXPLAIN as concrete methods for doing topic modeling, topics-criteria alignment, and self-explanation classification respectively. We assess the potential advantages of the proposed SENN4MCSA on a TripAdvisor data set. The user-orientated explanations are assessed by 4 sentiment criteria, which are often used to assess the reviews on this specific data set: room quality, value, location, and service. The empirical evidence suggests that SENN4MCSA can provide promising levels of predictive performance, and reliable user-orientated explanations.
Coastal erosion is a significant issue driven by various factors, including rising sea levels, human activities, and natural processes. This paper proposes a method to estimate the surface level using FastSAM, a convolutional neural network (CNN) designed for segmentation tasks. By using a point prompt, the model identifies the mask of the gauge and converts it to water surface level by counting the amount of the pixel height of the mask. The effectiveness of this method was evaluated using root mean square error (RMSE) and correlation coefficient. The proposed method was evaluated using data collected from a natural environment, Moonlight beach, Thailand. Compared to pressure sensor data, the proposed method achieve an RMSE of 0.7 cm and a correlation coefficient of 0.92. Additionally, when tested against manual observation of 3,000 images, the RMSE is 1.0 cm, with a correlation coefficient of 0.97. These indicate the promising potential of the proposed method in applying for coastal erosion monitoring systems.
The advantage of the interval priority weight vector estimated from a crisp pairwise comparison matrix over the estimated crisp priority weight vector has been demonstrated. Various methods for estimating interval priority weights have been proposed and compared by numerical experiments based on their performances to explore better estimation methods. The estimation methods based on minimum possible ranges have shown their good performances. However, the performances of those methods deteriorate to a certain extent when the widths of the assumed true interval priority weights decrease with decreasing the centers. Therefore, the exploration of better estimation methods has been continued. This paper proposes modified methods based on the minimum possible ranges by incorporating the center estimations in submodels. Numerical experiments about the estimation accuracy of interval priority weights and the accuracy of ordering alternatives are conducted. The results of the numerical experiments show that one of the proposed methods stably performs well.
Lung cancer remains a leading cause of cancer-related deaths, emphasizing the need for early and accurate detection. While traditional methods like computed tomography (CT) scans and X-rays are widely used, they rely on manual interpretation, leading to variability and errors. To address these limitations, this study evaluates the performance of three advanced deep learning models-Convolutional Neural Networks (CNNs), CSWin Transformers, and RoFormer-in detecting lung cancer from radiological images. The methodology includes a structured pipeline with preprocessing, model training, validation, and testing, using a three-way data split (training: 64
In this paper, we introduce a novel distance measure within the framework of Dempster-Shafer theory for quantifying the dissimilarity between two basic belief assignments. To demonstrate its effectiveness, the proposed method is compared with several widely used approaches using illustrative examples. Additionally, it is applied to uncertainty representation in numerical approximations of ODE solutions and in the analysis of biological data.
In this paper, a new approach for nonparametric estimation of probability density function is presented. It is based on Kolmogorov-Arnold Network that uses the activation functions on edges instead of on nodes. The experimental results show that the proposed method can obtain accuracy as high as the state-of-the-art methods while it has a compact architecture.
Linear fuzzy clustering-induced local PCA can have better interpretability than statistic-induced ones. The interpretability and stability have been demonstrated to be improved by considering cluster separation in the FCM clustering context. In this paper, with the goal of further improving the interpretability and stability of local PCA models, a novel linear fuzzy clustering model is proposed by introducing the cluster separation principle. Besides line-shape prototypes, 2-D plane-like linear prototypes and other linear varieties are adopted in conjunction with multi-dimensional local principal component scores. In order to consider mutual separation of linear cluster prototypes, the standard PCA criterion of point-wise information loss is replaced with the element-wise lower-rank approximation measure. Then, the novel clustering criterion is optimized by minimizing within-cluster errors and by maximizing intracluster PC score deviations and inter-cluster separation. Experimental results demonstrate that the proposed algorithm is useful for improving the initialization sensitivity of linear clustering with multi-dimensional prototypes.
Iris verification systems are known for their high accuracy in biometric identification; however, false rejections and false acceptances remain critical challenges, particularly when relying solely on iris features. In this study, we propose a score fusion approach that combines iris and periocular verification systems to enhance performance and reduce false negatives. Our proposed method integrates the iris verification system with periocular features processed through an autoencoder. The final verification decision is made through a score fusion module using a multi-layer perceptron (MLP) for iris verification and a support vector classifier (SVC) for the combined scores. Evaluating our approach on the CASIA-Iris V2 dataset, we observed improvements in recall, F1-score, and balanced accuracy compared to baseline methods, with a recall increase of 21.55
Wildfires present a severe risk to environmental sustainability, public health, and economic stability, especially in wildfire-prone areas, e.g., northern Thailand. This paper proposes a real-time wildfire-prone area monitoring and early warning system with a detection technique to address these challenges. First, we developed a web application that combines IoT cameras with a YOLOv5-based smoke detection model to monitor, detect, and notify users about potential wildfires. Second, we propose a two-stage smoke detection framework that leverages Gaussian filtering and a dual-stage YOLOv5 pipeline to reduce false predictions and improve detection accuracy. Using the FireSpot dataset comprising annotated images of early-stage wildfires, the system achieved a balanced accuracy of 98.62
In high-stability environments like server rooms, continuous monitoring is essential to promptly detect and report even minor environmental changes. Anticipating fluctuations and triggering timely alerts ensures proactive interventions to maintain optimal conditions. In this paper, we propose a system that employs minimal hardware, specifically an ESP SOC Devkit V1, to monitor the environmental quality of the network server room at HUFLIT University, Ho Chi Minh City. To detect potential environmental issues in multivariate time series data, we integrate a Variational Auto Encoder Bidirectional Long Short-Term Memory (VAE-Bi-LSTM) hybrid model as an unsupervised anomaly detection approach. As a result, our detection algorithm is capable of identifying issues that may not be immediately apparent, ensuring comprehensive monitoring. The system has been experimentally validated in real-world conditions, demonstrating not only stability and real-time interaction but also reliable long-term predictive accuracy, confirming its effectiveness in practical applications.
Annual Securities Reports (ASRs) prepared by Japanese companies are textual documents that investigate and report on their financial condition and potential risks. These documents include management policies, risks specific to companies or industries, and materials for management decisions, serving as critical resources for evaluating investment opportunities and financial health. By employing a clusteringbased approach, we aim to statistically identify commonalities and differences among companies and examine the relationships between riskrelated topics and various management indicators. Our analysis utilizes Fin-BERT, a domain-specific language model, and compares it with a general BERT to assess their ability to capture nuanced patterns in corporate risk descriptions. The results demonstrate that Fin-BERT effectively captures granular semantic information, particularly in financial contexts, which standard models might overlook. We observed a modest correlation between shifts in corporate risk descriptions and key management metrics, such as stock price fluctuations, indicating that risk narratives influence investor perceptions and pricing behaviors. These findings suggest opportunities to improve the integration of domain-specific language models into financial analysis and provide a foundation for future research to explore industry-specific variations and refine the methodologies for analyzing corporate risk descriptions.
This study investigates the factors influencing the development and implementation of social entrepreneurship education in Vietnamese universities, focusing on managing uncertainties in decision making and institutional strategies. Using a hybrid approach, the research integrates qualitative and quantitative methodologies to develop knowledge models, formulate hypotheses, and assess key determinants. The Partial Least Squares Structural Equation Modeling (PLS-SEM) framework is applied for data analysis and hypothesis testing. The findings reveal that cultural and ethical considerations, collaboration and networking, and institutional support significantly impact the effectiveness of social entrepreneurship education. This study highlights the role of uncertainty, such as variations in institutional priorities and stakeholder expectations, in shaping educational strategies. By addressing these uncertainties, the research contributes to a decision-support framework for integrating sustainable development goals into higher education curricula. These insights provide a pathway for Vietnamese universities and similar contexts to navigate the complexities of fostering social entrepreneurship education.
