Neural Computing and Applications

Paul, Tapas Kumar; Pal, Madhumangal

doi: 10.1007/s00521-026-12104-0pmid: N/A

In recent decades, there has been an increased focus on evaluating cloud computing vendors in India. Cloud computing (CC) facilitates the transformation of on-premises information and communication technology (ICT) resources through the Internet. Users can access these services via cyberspace through a compatible subscription on-demand. The foundational service model in CC is Infrastructure as a Service (IaaS). The abundance of Infrastructure as a Service vendors (IaaSVs) in the cloud computing industry poses a challenge for users in selecting the most suitable IaaSV. This research aims to identify a competent IaaSV in the public domain based on predefined qualitative and quantitative criteria, framing it as a multi-criteria group decision-making (MCGDM) problem. The criteria data are gathered in a Pythagorean fuzzy environment. The Removal Effect of Criteria (MEREC) method is utilized to assess objective weights (OWs) of the criteria, while subjective weights (SWs) are computed using the Stepwise Weight Assessment Ratio Analysis (SWARA) method. A combination of OWs and SWs yields the final criteria weights. Incorporating decision makers’ psychological behaviour under risks, a prospect theory-based extended Pythagorean fuzzy TODIM (Interactive Multi-criteria Decision Making) method is employed, leveraging Pythagorean fuzzy Jensen-Shannon Song distance to rank five selected IaaSVs. The proposed model’s validity is demonstrated through information from the g2 online database and a comparative study. Lastly, a sensitivity analysis underscores the robustness of the devised method.

Wushishi, Usman

doi: 10.1007/s00521-026-12220-xpmid: N/A

Industrial Internet of Things (IIoT) systems face complex threats from multiple concurrent attackers who employ diverse strategies under incomplete information about defender capabilities. Conventional game-theoretic cyber-deception models rely on fixed strategy equilibria and static beliefs, limiting adaptability to evolving attacks. Meanwhile, pure reinforcement learning approaches lack strategic reasoning about adversarial objectives. This paper introduces a hybrid architecture combining hypergame theory (a game-theoretic framework where players may hold different perceptions of the game structure) and deep reinforcement learning for adaptive IIoT defense. Beliefs are updated using Random Forest classifiers trained on the CIC-IIoT 2025 dataset (500,000+ samples, 71 features, 8 attack categories), achieving 98.2% classification accuracy. The defender selects strategies via hypergame Nash equilibrium across four defense bundles and eight attacker strategies, with payoff matrices grounded in ML-derived detection rates. A Proximal Policy Optimization (PPO) meta-learner refines these equilibrium strategies through online interaction, balancing game-theoretic structure with adaptive learning. Experiments over 200 episodes show that our framework achieves 71.8% detection rate with 3.8% false positives against two simultaneous attackers, compared to static hypergame solutions (64.9% DR, 5.3% FPR) and non-learning baselines (35–56% DR, 6–9% FPR). To the best of our knowledge, this is one of the first approaches to integrate game theory, machine learning, and reinforcement learning for proactive defense under belief uncertainty.

Atef, Markus; Gabr, Menna Ibrahim; Seoud, Wafaa; Ouf, Shimaa

doi: 10.1007/s00521-026-12226-5pmid: N/A

Peer-to-peer (P2P) lending has increased significantly during the past few years on a global scale. However, there are several challenges associated with P2P lending’s rapid rise. The major challenges are imbalanced datasets, which make machine learning difficult, an excessive number of features, and low-performing classification algorithms. Furthermore, machine learning models face another complex challenge referred to as the black-box problem. To address these challenges, an innovative approach was developed by first applying Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance in the Bondora dataset, followed by the implementation of multiple feature selection techniques: Chi-Square (filter), Sequential Backward Selection (SBS) (wrapper), and embedded methods such as Random Forest (RF), Gradient Boosting Machine (GBM), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost). A range of classifiers, linear (Logistic Regression (LR)), non-linear (Support Vector Machine (SVM), Naive Bayes (NB), and tree-based models (Decision Tree (DT), RF, Adaptive Boosting (AdaBoost), CatBoost), were then used to predict loan defaults. The top-performing models were integrated into various stacking ensembles using GBM, Extreme Gradient Boosting (XGBoost), and LightGBM as meta-learners to enhance predictive accuracy. The results declared that LightGBM exhibited an outstanding performance with accuracy, F-score, and Area Under the Curve (AUC) values of 0.981, 0.980, and 0.994, respectively, showing better performance than that reported in the literature. Explainable models were employed to interpret predictions and enhance user trust. Specifically, the LightGBM stacking model was combined with the Local Interpretable Model-agnostic Explanations (LIME) framework to provide interpretable insights into its prediction results.

Patel, Ronak R.; Patel, Miral

doi: 10.1007/s00521-026-12250-5pmid: N/A

Glioblastoma is a high-grade brain tumor that causes a high risk of death. Early detection of such tumors helps to improve human life. MRI scans are one of the most popular diagnostic reports to identify such kind of complex diseases. Such advanced diagnosis reports help to identify tumor size, location, and aggressiveness. The proposed architecture uses a hybrid bidirectional approach to share the feature with prior and domain branches. Domain branch focuses on the volumetric context for enhancing the boundary of tumor. Prior branch works on 2D and 3D fusion to identify spatial information for highly affected cells. Identification of sharp boundaries of the tumor is challenging. Based on the intensity of all modalities, the Residual Feature Interaction Network (RFIN) focuses on the non-enhancing regions. On the other end, based on the spatial information Domain Knowledge Interaction Network (DKIN) component focuses on WT. RFIN and DKIN act in a bi-directional manner for better identification of the region. The proposed architecture gives remarkable results based on benchmark datasets BraTS2019 and BraTS2020. For evaluation of the proposed approach, the mean DSC is considered, and the results are 0.8184, 0.8735, and 0.8902 for ET, WT, and TC, respectively.

Ahmmed, Md Ashik; Rabbi, Rashel Mahmud; Shafiuzzaman, Md; Ahamed, Md. Faysal; Nahiduzzaman, Md; Chowdhury, Muhammad E.H.

doi: 10.1007/s00521-026-12192-ypmid: N/A

Satellite image classification has many important applications that play a crucial role in the areas of urban planning, agriculture, as well as environmental monitoring. Nevertheless, the high accuracy and interpretability of deep learning models with such complex datasets is still a big challenge. To solve this, a new hybrid deep-learning architecture, MRCL-ELM is proposed to improve the performance of satellite image classification. The model uses EfficientNetB0 as its building block in terms of ability to learn rich features in an efficient manner with optimization of the network depth and size to minimize memory and processing requirements. It combines the Multi-Residual Convolutional (MRC) networks to learn spatial features robustly with the aid of multiple residual paths, in each block, in MRC, to enhance the learning of features and gradient flow. It uses a Long Short-Term Memory (LSTM) time series modeling layer, and an Extreme Learning Machine (ELM) to quickly and non-iteratively classify data and is therefore lightweight, accurate, and more scalable than other common deep learning architectures. To enhance the interpretability of the proposed model, Local Interpretable Model-agnostic Explanations (LIME) explains individual predictions by testing small variations in the input, whereas SHapley Additive exPlanations (SHAP) provides feature importance scores throughout the model, along with improving model interpretability and trust. The proposed model provides the highest possible results, with 98.33% accuracy on the EuroSAT dataset and 98.10% accuracy on the UC Merced Land Use dataset, being higher than the use of existing Convolutional Neural Networks (CNN) and transformer-based techniques. The training using fixed random seeds and 5-fold cross-validation is used to ensure robustness. Lastly, MRCL-ELM was implemented as a real-time web-based application and tested with real-life Google Maps imagery, and thus needs real-time, precise, and interpretable satellite image classification for the end-users.

Bakar, Mohamad Ariffin Abu ; Ab Ghani, Ahmad Termimi; Abdullah, Mohd Lazim; Yussof, Fatin Nadiah Mohamed

doi: 10.1007/s00521-026-12193-xpmid: N/A

The ability to solve mathematics problems results from the relationship between critical cognitive skills and behavioral influences. This ability is important for ensuring academic success and can be applied in the real world. However, conventional assessment methods often fail to capture the multifaceted and uncertain nature of these abilities. Therefore, this study introduces a novel Single-Valued Neutrosophic (SVN) Entropy and Decision-Making Trial and Evaluation Laboratory (DEMATEL) based hierarchical Adaptive Neuro-Fuzzy Inference System (ANFIS) model to predict students’ mathematics problem-solving ability. The proposed model integrates SVN Entropy for attribute weighting in data preparation and SVN DEMATEL for the hierarchical structuring of its feature engineering modifications, thereby overcoming limitations commonly observed in traditional ANFIS models, especially in handling indeterminacy, vagueness, and complex causal relationships. The proposed model demonstrates superior predictive performance compared to state-of-the-art machine learning models due to its enhanced capability to manage uncertainty and cognitive, behavioral interactions. The model proves the credibility and reliability of the modifications implemented by recording an increase in prediction accuracy and thus reducing the error in it. Performance metrics show that the model configuration with three triangular membership functions is the most consistent by yielding improvements in root mean square error (RMSE) of 61.93%, mean absolute error (MAE) of 15.83%, and R² of 64.93% compared to the traditional ANFIS model. These results highlight the advantages and novelty of the proposed hybrid neutrosophic, fuzzy framework and contribute valuable insights to the mathematical sciences, particularly in using soft computing techniques to improve the predictive power of assessment models in education. This work further supports the development of a more synergistic framework for cognitive and behavioral assessment within complex humanitarian systems.

Conti, Edoardo; Rosati, Riccardo; Federici, Lorenzo; Mancini, Adriano; Fiorentino, Maria Chiara

doi: 10.1007/s00521-026-12210-zpmid: N/A

This study provides the first comprehensive evaluation of foundation models in fetal ultrasound (US) imaging under low inter-class variability conditions. While recent vision foundation models such as DINOv3 have shown remarkable transferability across medical domains, their ability to discriminate anatomically similar structures has not been systematically investigated. We address this gap by focusing on fetal brain standard planes–transthalamic (TT), transventricular (TV), and transcerebellar (TC)–which exhibit highly overlapping anatomical features and pose a critical challenge for reliable biometric assessment. To ensure a fair and reproducible evaluation, all publicly available fetal ultrasound datasets were curated and aggregated into a unified multicenter benchmark, FetalUS-188K, comprising more than 188,000 annotated images from heterogeneous acquisition settings. DINOv3 was pretrained in a self-supervised manner to learn ultrasound-aware representations. The learned features were then evaluated through standardized adaptation protocols, including linear probing with frozen backbone and full fine-tuning, under two initialization schemes: (i) pretraining on FetalUS-188K and (ii) initialization from natural-image DINOv3 weights. Models pretrained on fetal ultrasound data consistently outperformed those initialized on natural images, yielding weighted F1-score improvements of up to 21% (0.73 vs. 0.52 for ViT-B/16). This domain-adaptive pretraining proved critical for resolving low-margin class boundaries; while natural-image weights led to a representational collapse on the TV plane (F1-score \documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\le$$\end{document} 16%), our approach preserved the subtle echogenic and structural cues necessary for its accurate discrimination. These results demonstrate that while generic foundation models fail to generalize under low inter-class variability, domain-specific pretraining is a technical prerequisite for achieving the robust and clinically reliable representations required for fetal brain biometric assessment.

de Cássia Lobato Soares, Amanda; Blanco, Claudio; de Mendonça, Leonardo Melo; da Silva Cruz, Josias

doi: 10.1007/s00521-026-12146-4pmid: N/A

This study aims to simulate streamflow using machine learning in a basin located in the Brazilian Amazon under two future climate scenarios from CMIP6, and analyze the impacts of climate change on streamflow until 2100 through trend analysis. A Nonlinear Auto Regressive Recurrent Neural Network with Exogenous Inputs (NARX) was trained to project streamflow under the SSP2-4.5 and SSP2-4.5 and SSP5-8.5 scenarios. Precipitation projected by the Global Circulation Models (GCMs) GFDL-ESM4, FGOALS-g3 e CESM2 was used as input to the model. The maximum streamflow simulated by NARX model in the reference period were underestimated. This underestimation was attributed to a systematic error in the precipitation projected by the GCMs, characterized by the delay in the peak of maximum. Therefore, the empirical quantile mapping EQM method was applied to correct the bias in the simulated streamflow using observed data from the reference period. The Mann-Kendall method was used to analyze future streamflow trends. The results show that the overall performance of the simulations was classified as good, with Kling-Gupta (KGE) values ranging from 0.73 to 0.77 in both scenarios. This performance was also reflected in the cumulative distribution function (CDF) curves of streamflow, which showed good agreement with the observed streamflow distribution. The Mann–Kendall test results for the simulated streamflow under the SSP2-4.5 scenario indicate stability in the streamflow regime. In contrast, for SSP5-8.5, a stronger signal appears mainly in GFDL-ESM4, which shows a significant decreasing trend (p = 0.0071) with Sen’s slope = − 0.313, indicating possible streamflow reduction. FGOALS-g3 (p = 0.5120) and CESM2 (p = 0.1527) showed no significant trends, although their slopes suggest opposite tendencies (0.0566 and − 0.1533, respectively). The MME also showed no significant trend (p = 0.1029), but its negative slope (− 0.1488) suggests a general tendency toward decreasing streamflow.

Sedghnejad, Nadia; Nozari, Hamed; Marofi, Safar

doi: 10.1007/s00521-026-12106-ypmid: N/A

Predicting runoff is a crucial aspect of preventing floods and droughts, safeguarding reservoirs, and managing water resources. The exceptional accuracy of artificial intelligence models in predicting hydrological parameters has attracted considerable attention from researchers. Consequently, this study employed Support Vector Machine models optimized by Simulated Annealing (SVM-SA) and Particle Swarm Optimization algorithms (SVM-PSO), as well as Linear Regression (LR) and Multiple Linear Regression (MLR) models utilizing Single Input-Single Output (SISO) and Multiple Input-Single Output (MISO) patterns to predict monthly runoff at 25 hydrometric stations in the Karkheh basin, located in Iran. For this purpose, the statistical data were divided into 80% calibration and 20% validation. The statistical analysis of the results utilized the coefficient of determination (R2), Standard Error (SE), and Root Mean Square Error (RMSE), Nash–Sutcliffe Efficiency (NSE), Kling Gupta Efficiency (KGE), and Percent Bias (PBIAS) as indicators. Among the SISO models assessed, the LR model demonstrated the most favorable performance, achieving an average Standard Error of 0.41, a R² and NSE of 0.87, a PBIAS close to zero, and a KGE of 0.90. These metrics indicate a robust capability to replicate monthly runoff patterns effectively. The SVM-PSO and SVM-SA models also exhibited commendable performance, with SE values ranging from approximately 0.41 to 0.42 and NSE values between 0.87 and 0.88. The KGE scores for these models were 0.88 and 0.87, respectively. It is noteworthy that SVM-SA exhibited narrower prediction intervals and lower sensitivity, as indicated by an Average Relative Interval Length (ARIL) of 0.07, whereas SVM-PSO produced wider prediction intervals with an ARIL of 2.16. In the MISO configuration, the MLR model outperformed all other models, recording the lowest SE of 0.39, a PBIAS of 0, a KGE of 0.90, and the most compact ARIL of 0.03. These results reflect high accuracy, stability, and reliability in predicting monthly runoff. The SVM-SA model demonstrated balanced performance with a higher KGE of 0.93, while the SVM-PSO model displayed a lower KGE of 0.88 and broader uncertainty coverage, albeit with wider prediction intervals (ARIL = 2.46). The SVM-PSO model provided broader uncertainty coverage but at the expense of wider intervals, while the SVM-SA model showed moderate performance with acceptable sensitivity and prediction accuracy. Overall, the MLR model under the MISO structure proved to be the most effective approach for monthly runoff prediction due to its combination of high predictive accuracy, computational simplicity, stable sensitivity profile, and reliable uncertainty quantification.

Sousa, João; Henriques, Roberto

doi: 10.1007/s00521-026-12209-6pmid: N/A

Time-series forecasting of multiple related sequences presents unique challenges due to the complex interplay between individual series characteristics and global patterns. We present T2f, a forecasting method combining ensemble learning with an actor-critic architecture based on the Twin Delayed Deep Deterministic algorithm (TD3). T2f balances local and global patterns through both its architecture and learning approaches, integrating transformer-based pattern recognition with reinforcement learning for dynamic model selection. Our method incorporates temporal attention mechanisms and context-aware error measurement, aligning forecasting objectives with practical decision-making priorities. Comprehensive ablation studies demonstrate that T2f’s components provide synergistic benefits: the TD3-based optimizer contributes 18.8% error reduction over static weighting, while temporal attention adds 8.0% improvement, with the integrated system outperforming simple ensemble baselines by over 20%. Experimental results across five diverse datasets indicate T2f reduced mean absolute error by over 30% compared to statistical models and achieved up to 40% better performance on context-weighted metrics than competing approaches. While specialized models occasionally outperformed T2f on highly regular patterns, it consistently showed superior adaptability to contextual weights with faster convergence, typically reaching near-optimal performance within 25 epochs compared to 40+ for alternative methods, particularly on datasets with complex temporal dynamics.