Foundations of Computing and Decision Sciences Feed

A Deep Learning Approach to Classifying Software Requirements: The Application of Transformer-Based Ensemble Learning and Attention-Based Fusion

Tue, 17 Mar 2026 00:00:00 GMT

The functional requirements (FRs) classification in software requirements classification (SRC) is a difficult task due to class imbalance, fine-grained subcategories, and semantic complexities. Existing Machine Learning (ML) and Deep Learning (DL) models often rely on handcrafted characteristics or overlook the contextual meaning. This work presents a novel hybrid ensemble framework that refines three pre-trained transformers (BERT, DistilBERT, and RoBERTa) and combines them using two mechanisms: (1) an Attention-Based Fusion Mechanism that dynamically selects the most contextually relevant transformer for each instance, and (2) an Accuracy-Per-Class Weighted Ensemble that assigns weights based on per-class validation accuracy. Tested on multiple datasets, the approach outperformed single-transformer and DL models (CNN, LSTM, BiLSTM, and GRU) by a large margin (p < 0.001), achieving 95% accuracy and 0.94 F1-score. To the best of our knowledge, this is the first study to combine attention fusion and transformer-based ensembles for SRC, establishing a new standard for SRC.

The Potential of Prospect Theory under Inductive Reasoning

Tue, 17 Mar 2026 00:00:00 GMT

The TODIM method was the first Multiple Criteria Decision Making (MCDM) approach to incorporate the advances of judgment and decision theory based on the findings of prospect theory. Since its advent, several studies have proposed new methods for the same purpose, such as Behavioral TOPSIS. Additionally, new versions of the original TODIM method have been developed to enhance its adherence to prospect theory. Among these approaches, the ExpTODIM version demonstrated high accuracy in predictions under deductive reasoning scenarios. In this research, an inductive reasoning analysis of the potential of ExpTODIM and Behavioral TOPSIS is conducted. The results are then compared with a native inductive reasoning method, namely multinomial regression analysis. It was found that ExpTODIM demonstrated strong potential for applications under an inductive reasoning perspective, which supports the application of prospect theory in tasks such as autonomous classification, a key area in artificial intelligence studies.

Influence of Graphical Representation Type on Tessellated Geometry Classification

Tue, 17 Mar 2026 00:00:00 GMT

This research investigates the impact of graphical representation on the accuracy of three-dimensional shape classification. 3D models, both scanned and modeled, are used in engineering, computer graphics and scientific data visualization. Various approaches are adopted in these fields to visually represent 3D geometry, utilizing solutions such as OpenGL and Direct3D, each with its distinct goal of achieving either real-time manipulation or photorealism. The purpose of this research was to determine the most effective graphical representation for categorizing mechanical components with a high degree of geometric similarity, such as beams and rods. The study examined various image representations and their combinations, obtained through the adjustment of rendering parameters and image compositing. In an effort to improve classification accuracy, novel techniques for addressing image recognition issues were developed and tested against commonly used image representation methods. This innovative approach proposed in the paper led to a 48% reduction in classification errors.

Machine Learning for Multi Objective Convex Separable Programming (MOCSP) with Aggregation of Linear Approximations and Portfolio Optimization

Tue, 17 Mar 2026 00:00:00 GMT

A novel technique is developed for nonlinear optimization problem which is convex, separable and having multiple objective functions. In the development of the model all the objectives and the constraints of the multi objective model are linearly approximated over suitable intervals. The linear approximations are then aggregated to account for the original problem. The developed technique has been utilized for portfolio optimization problem. Firstly, the minimum variance model has been formulated and solved with machine leaning techniques. Secondly, the risk aversion model has been formulated and solved. The results obtained are combined into a multi objective framework of convex separable programming problem. All the three problems have been solved with the help of the XGBoost, neural network, and decision forest regression models. The renowned Python machine libraries of scikit-learn and keras have been utilized. The results identified portfolios that can return more financial benefits to the investors while investing in the capital market. The results of the proposed MOCSP approach are 22.5% improved in case of risk aversion model. Additionally, 17% improvement has been recorded in case of the minimum risk model. The MAE and RMSE for both XGBoost and decision forest regression have a frail value 0.0001. MAE and RMSE for the neural network regression have been recorded 1% and 2%, respectively. Both Accuracy and F1 score for XGBoost are 91%, for neural network regression are 98%, and for decision forest are 92%, respectively.

MOPOA: A New Multi-Objective Pufferfish Optimization Algorithm

Tue, 17 Mar 2026 00:00:00 GMT

Multi-objective optimization problems (MOPs) pose significant challenges due to the presence of multiple conflicting objectives. This paper introduces MOPOA, a novel Multi-Objective Pufferfish Optimization Algorithm inspired by the defensive behaviors of pufferfish in nature. MOPOA extends the original single-objective POA by incorporating Pareto dominance, an external archive for preserving non-dominated solutions, and a crowding distance mechanism to maintain solution diversity. The algorithm balances exploration and exploitation through biologically inspired phases simulating predator-prey interactions. To evaluate MOPOA’s performance, it was benchmarked against several state-of-the-art algorithms, including NSGA-III, MOPSO, MODA, and MOFDO, on two well-known test suites: the ZDT and CEC-2019 multi-objective functions. Results indicate that MOPOA not only achieves superior convergence to the Pareto front but also maintains high diversity and robustness across diverse optimization scenarios. These findings position MOPOA as a powerful and adaptive tool for solving complex real-world multi-objective problems.

f-divergence Analysis of Generative Adversarial Network

Mon, 08 Dec 2025 00:00:00 GMT

We aim to establish estimation bounds for various divergences, including total variation, Kullback-Leibler (KL) divergence, Hellinger divergence, and Pearson χ2 divergence, within the GAN estimator. We derive an inequality based on empirical and population objective functions of the GAN model, achieving almost surely convergence rates. Subsequently, this inequality was employed to derive estimation bounds for total variation, Kullback-Leibler (KL) divergence, Hellinger divergence, and Pearson χ2 divergence, leading to almost surely convergence rates and differences between the expected outputs of the discriminator on real data and generated data. Our study demonstrates better results compared to some existing ones, which are a specific case of the general objective function.

DermoAI-CNN: Leveraging GANs, Mask R-CNN, and Attention Mechanisms for Enhanced Skin Disease Analysis

Mon, 08 Dec 2025 00:00:00 GMT

The study aims to develop an effective and efficient deep learning model for detecting skin diseases, as skin diseases rank as the world’s number one health problem. Besides, cancers and dermatological anomalies should be diagnosed at an early stage, so that subsequent treatment can be efficient and complication-free. The existing methods of diagnosis are associated with lower precision and, in most cases, are inefficient, which can be attributed to the lack of effective data augmentation, segmentation techniques, and improved feature extraction. In this paper, a general framework is introduced that uses Generative Adversarial Networks for data augmentation, Mask R-CNN for precise segmentation, and a tailored multilayer Convolutional Neural Network with an attention mechanism incorporated into it to classify 23 skin disease classes using 25,250 images, among them 5,750 generated by GAN, to balance underrepresented classes. The accuracy attained was 97.30%, which was much better than that reported in earlier studies, which ranged from 85 to 92. The metrics, including an accuracy of 95.65%, a recall of 97.09%, and an F1-score of 96.98%, were used to assess the system’s performance in classifying invisible dermatological images. The scalable system provides explanations that support real-time diagnosis, preventing delays and acute health costs. The findings fully fulfil the capabilities of deep learning in dermatology, as the initial diagnosis of the skin disease is accurate, accessible and efficient.

Leveraging Artificial Intelligence for Cyanobacterial Bloom Prediction: A Hybrid Deep Learning and Generative Adversarial Network Framework for Accurate Forecasting and Proactive Management

Mon, 08 Dec 2025 00:00:00 GMT

This study presents an Artificial Intelligence-based system designed to predict cyanobacterial harmful algal blooms (CyanoHABs). The system utilizes Long Short-Term Memory (LSTM) networks to predict the timing of bloom occurrences and One-Dimensional Convolutional Neural Networks (1D-CNNs) to estimate cyanobacterial density. Additionally, Generative Adversarial Networks (GANs) are employed for data augmentation to enrich the database. The system’s performance was validated using the Algerian Mexa database, achieving an R-squared (R2) value of 98% and a root mean square error (RMSE) of 9% for cyanobacterial density prediction, and an R-squared value of 88% with a root mean square error of 31% for bloom timing prediction. These results highlight the system’s robust predictive capabilities, enabling proactive monitoring and management of CyanoHABs to mitigate their adverse impacts on health and the environment.

A Transparent AI-Driven Multiclass Decision Support System for Thyroid Risk Prediction Using Machine Learning and Deep Learning Approaches

Mon, 08 Dec 2025 00:00:00 GMT

Early and accurate diagnosis of thyroid disorders is essential due to their prevalence and health impact. To enhance interpretability in clinical settings, we propose a comprehensive workflow for transparent thyroid disease prediction using a multiclass classification problem with five diagnostic categories. A dataset of 9172 samples with 31 features was used to train various machine and deep learning models. A dual-layered framework combining Feature Selection (ETC, MI, RFE) and Explainable AI (SHAP, LIME) improved performance and transparency. Gradient Boosting achieved the highest accuracy (0.97). SHAP explained global feature influence, while LIME clarified individual predictions. Our approach supports interpretable, reliable AI-based diagnostic tools for thyroid disorder classification.

Quantifying the Digital Divide: A Data-Driven Approach with EFA, Shannon Entropy and Sensitivity Testing

Mon, 08 Dec 2025 00:00:00 GMT

In the digital world, inequalities in digital participation often arise as a consequence of the uneven distribution of digital resources. Therefore, the objective of this study is to develop a digital divide index (DDI) that facilitates the temporal and spatial examination of disparities in the availability and quality of internet access, internet use and outcomes. Empirical data for 27 European countries were collected from online available datasets for 2014 and 2022. Exploratory factor analysis (EFA) was used to identify factors in DDI, while the weight values were calculated using the Shannon entropy method. The average difference in DDI scores indicates a spread of the digital divide facilitated by Finland, Latvia, Estonia, Germany, Poland and Slovakia. States in Northern and Central Europe like Norway, Austria and the Czech Republic are at the forefront of achieving digital equality. Spatial analysis reveals that Benelux countries along with Scandinavian countries show the highest levels of DDI, while Southern Europe lags behind. Sensitivity testing results show a stable index structure with variations in the importance of factors not significantly affecting the ranking results.

Improving Machine Learning-Based Robot Self-Collision Checking with Input Positional Encoding

Thu, 21 Aug 2025 00:00:00 GMT

This manuscript investigates the integration of positional encoding – a technique widely used in computer graphics – into the input vector of a binary classification model for self-collision detection. The results demonstrate the benefits of incorporating positional encoding, which enhances classification accuracy by enabling the model to better capture high-frequency variations, leading to a more detailed and precise representation of complex collision patterns. The manuscript shows that machine learning-based techniques, such as lightweight multilayer perceptrons (MLPs) operating in a low-dimensional feature space, offer a faster alternative for collision checking than traditional methods that rely on geometric approaches, such as triangle-to-triangle intersection tests and Bounding Volume Hierarchies (BVH) for mesh-based models.

A Mixed Integer Programming Approach to the Rechargeable Rover Routing Problem on Mars

Thu, 21 Aug 2025 00:00:00 GMT

In this paper, we introduce a novel variant of the Vehicle Routing Problem (VRP), the Rechargeable Rover Routing Problem (RRRP), which addresses the routing of energy-constrained autonomous electric rovers for Martian missions. We formulate a graphbased representation of the problem and propose an initial formulation as a mixed-integer non-linear program (MINLP). To enhance computational efficiency, we demonstrate how the model can be linearized. The resulting mixed integer linear model is evaluated on small-scale test cases, and its computational complexity is analyzed for larger problems with up to 30 Points of Interest (PoIs). Our experiments show that the problem can be solved to optimality for problem sizes anticipated in upcoming Mars expeditions. However, for future missions involving swarms of rovers, the development of more efficient heuristic or approximation algorithms will be necessary.

Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds

Thu, 21 Aug 2025 00:00:00 GMT

Semantic segmentation is important for robots navigating with 3D LiDARs, but the generation of training datasets requires tedious manual effort. In this paper, we introduce a set of strategies to efficiently generate large datasets by combining real and synthetic data samples. More specifically, the method populates recorded empty scenes with navigation-relevant obstacles generated synthetically, thus combining two domains: real life and synthetic. Our approach requires no manual annotation, no detailed knowledge about actual data feature distribution, and no real-life data of objects of interest. We validate the proposed method in the underground parking scenario and compare it with available open-source datasets. The experiments show superiority to the off-the-shelf datasets containing similar data characteristics but also highlight the difficulty of achieving the level of manually annotated datasets. We also show that combining generated and annotated data improves the performance visibly, especially for cases with rare occurrences of objects of interest. Our solution is suitable for direct application in robotic systems.

Genetic Algorithm Approach to a Concurrent Real-Time Optimization Problem in the Embedded System Design Process

Thu, 21 Aug 2025 00:00:00 GMT

In this paper, we present a genetic algorithm for a concurrent real-time optimization problem occurring in the embedded system design process. The problem consists of two concurrent phases, each impacting the other in real time. In the first phase, parameters are selected for optimization, and in the second, the parameters are optimized and their choice is validated in real time. During the implementation of the embedded system, unexpected situations can arise, each of which can be solved in many ways; each way, in turn, may require the execution of different unexpected tasks. However, identifying the optimal path to follow is significantly challenging. Furthermore, some of the proposed solutions to the problem may not yield appropriate results. The proposed algorithm generates a certain number of individuals and evolves them using genetic operators, performing the proper optimization and comparing the results.

Preface to the Issue on Recent Developments in AI-Based Robotics: A Polish Perspective

Thu, 21 Aug 2025 00:00:00 GMT

Overlapping Box Suppression and Merging Algorithms for Window-Based Object Detection

Thu, 21 Aug 2025 00:00:00 GMT

In this manuscript, we extend the Overlapping Box Suppression (OBS) algorithm, a novel approach designed to enhance window-based object detection systems by reducing false-positive detections. While window-based methods are commonly used for small object detection, they often face challenges due to partially visible objects and intersecting detection windows. To address this, the proposed OBS algorithm uses the detection window coordinates to effectively filter out redundant partial detections, improving detection quality. Additionally, we introduce a novel Overlapping Box Merging (OBM) algorithm, which further refines detection results by combining partial detections into a single, more accurate detection. Together, OBS and OBM offer a robust solution for handling overlapping and fragmented detections. We evaluate this combined global filtering block on sequences from the SeaDronesSee dataset, demonstrating superior performance across multiple object detection metrics compared to traditional NMS-based filtering methods.

Algorithm for Determining the Sum Formula of Metabolites from Mass Spectrometry Spectra

Tue, 10 Jun 2025 00:00:00 GMT

This work addresses the challenge of determining chemical sum formulas from mass spectrometry data through a three-fold approach: establishing a formal problem framework, analyzing its computational complexity, and developing an approximate algorithm that synergizes reference database integration with functional group property analysis to enable accurate compound identification and structural characterization. By bridging theoretical foundations in computational complexity with practical solutions for chemical structure elucidation, the proposed methodology advances analytical capabilities in mass spectrometry data interpretation.

A Novel Weighted Preference Relation Approach to Detect Outliers in Multi-Criteria Decision Aid Context

Tue, 10 Jun 2025 00:00:00 GMT

The problem of outlier detection in the multicriteria decision aid (MCDA) field has not been extensively explored in the current literature. This study presents a novel approach to tackle this challenge, based on two key concepts. Firstly, the degree of importance of a preference relation, which utilizes multicriteria preference indices (derived from the PROMETHEE method) to assess the significance level of a preference relation. Secondly, the similarity of alternatives, which uses the degree of importance to evaluate how similar each alternative is to the others. Based on the distribution of these similarities, outliers are identified using either the Interquartile Range (IQR) method or the Standard Deviations (SD) method. The proposed approach is applied to two real-world scenarios: the world happiness ranking problem and the human development index problem.

Repairing ETL Processes using Extended Relational Algebra

Tue, 10 Jun 2025 00:00:00 GMT

In a data warehouse architecture, heterogeneous and distributed data sources (DSs) are integrated by means of an extract-transform-load (ETL) layer, which runs integration processes (a.k.a. ETL processes). This layer is not static, since DSs being integrated change their schemas in time. A DS schema change impacts ETL processes, which typically stop working and need to be re-designed (i.e., repaired). Our overall goal is to repair automatically these ETL processes that were affected by DS schema changes. In this paper we focus on ETL processes specified by extended relational algebra, since relational data warehouses are among the most popular for business applications. For such a processes, we contribute a repair method. The method uses a rule engine that maps a possible DS schema change with: (1) an ETL operation on the changed schema element and with (2) a repair rule applicable if a DS schema element is changed. Based on this mapping, when a DS schema change occurs, our solution allows to apply adequate ETL rules to repair the affected ETL processes.

Determination of Criteria Priorities and Interactions in ChatGPT by a Fuzzy Model

Tue, 10 Jun 2025 00:00:00 GMT

Today, technology is advancing much more and faster than in the past. The technological revolution is progressing so fast that people can now communicate with preprogrammed computers through Artificial Intelligence (AI). This study aims to evaluate the priorities and interactions of the criteria that stand out in the use of ChatGPT-4, using Fuzzy-Analytical Hierarchy Process (F-AHP) and Fuzzy-DEMATEL (F-DEMATEL) methods. Eight critical criteria in the use of ChatGPT are determined by the expert opinions obtained from the focus group. Among the criteria, this study found that reliability 23.1% and security 21.2% were the most important criteria by F-AHP calculations. This study also found that the criteria of appearance and functionality were generally influencing criteria, and the criteria of speed was generally influenced by other criteria.