Future Technology

A framework for integrating artificial neural networks and finite element analysis for shear strength prediction in unreinforced RC beam–column joints

Ranim Maatouq, Rouya Hdeib — Sat, 11 Jul 2026 00:00:00 +0000

Beam-Column joints (BCJs) are critical components in reinforced concrete (RC) buildings. These members experience excessive stress during seismic events, often resulting in catastrophic failures, particularly in RC buildings constructed prior to the introduction of seismic design provisions and lacking reinforcement in the BCJ zone. The study presents a machine learning framework to predict the shear strength of unreinforced BCJs using 7 input parameters. This study developed an Artificial Neural Network-Finite Element Analysis hybrid model (ANN-FEA-13), trained, validated, and tested on 4320 samples of BCJ failure generated through nonlinear analysis in ABAQUS. The data was divided into training (70%), testing (15%), and validation (15%) sets. The ANN-FEA-13 model achieved high prediction accuracy (R = 0.962) and was compared with experimental data from literature and the ACI 318 code, showing superior performance. The results were promising and demonstrated the effectiveness of the developed data-driven ANN-FEA-13 framework, which reliably predicts BCJ failure and supports ongoing efforts in resilience-based assessment and retrofitting of aging RC structures in seismic regions.

Multi-agent generative AI ecosystems for cyber-physical systems in Industry 5.0

Kanaparthi Anil Kumar, K Hemachandran — Thu, 16 Jul 2026 00:00:00 +0000

Cyber-Physical Systems (CPSs) are increasingly being implemented in critical industrial infrastructure, where complexity and interdependence are rising, posing significant cyber and operational risks. Traditional anomaly detection algorithms can be ineffective in capturing temporal dynamics, relational dependencies, and interpretable response requirements. The present paper proposes a multi-agent generative AI system to detect CPS anomalies and provide decision support by combining temporal feature encoding, relational modeling as graphs, supervised learning, and reasoning with an LLM. The architecture consists of detection, diagnosis, planning, governance, and human-in-the-loop validation agents. The framework is tested on the SWaT benchmark data. Findings indicate that the Autoencoder, LSTM, and 1D-CNN are more effective in terms of raw detection metrics, whereas the Random Forest provides more interpretable and agent-readable evidence to support the post-detection decision. Analysis of features and sensor family suggests the relevance of relational dependencies in characterizing anomalies. The multi-agent layer converts selected anomaly predictions into context-dependent explanations and governance-filtered recommendations to aid operator review, response planning, and process resilience. Overall, the framework supports transparent and human-supervised CPS anomaly management aligned with Industry 5.0 principles.

AI-driven liability assessment framework for autonomous systems integrating machine learning with EU and Indian regulatory models

Fri, 17 Jul 2026 00:00:00 +0000

Autonomous systems pose liability issues when automated decisions contribute to harm, failure, or non-compliance with regulations. Current legal AI studies primarily focus on classification, retrieval, question answering, and judicial prediction, and little has been done to translate legal and regulatory texts into structured signals that convey liability-relevant information. This research paper proposes a decision-support model for categorizing liability cues under European Union and Indian law. The source datasets were MultiEURLEX and IndicLegalQA, and a rule-based proxy mapping was developed to assign five labels: damage, defect, liability, regulation, and risk. This model is based on word-level and character-level TF-IDF, jurisdiction encoding, and a One-vs-Rest Linear Support Vector Machine classifier for multi-label classification. The final best configuration had an overall 91.33% subset accuracy, 96.08% Micro-F1, 87.81% Macro-F1, and 0.0201 Hamming Loss, compared to the best configurations of Logistic regression, Linear SVM, and Rand Forest. The framework aids initial legal and regulatory analysis and does not decide liability on doctrinal grounds or substitute for proficient legal analysis.

Evaluating human-like chatbot interaction in customer service: integrating anthropomorphism and conversational capability

Warjiyono Warjiyono , Hadiyanto Hadiyanto, Dinar Mutiara Kusumo Nugraheni — Sat, 18 Jul 2026 00:00:00 +0000

The development of artificial intelligence (AI)-based technology has driven the development of rule-based chatbots into AI-based chatbots. This development is used by various companies to use AI chatbots as the frontline of customer service. Meeting user expectations for AI chatbots requires an evaluation model to measure chatbot quality. Based on the literature, chatbot evaluation models still focus solely on technical quality, even as AI chatbots are evolving towards human-like services. This research will explore the integration of the technical dimensions of information system success with a new dimension, namely Human-Like Chatbot Interaction (HLCI) as a higher-order construct of anthropomorphism and conversational capability. This quantitative study uses an explanatory research approach. Data were collected through a cross-sectional survey of users of the Veronika chatbot, owned by Telkomsel, an Indonesian telecommunications company. A total of 213 respondents had data ready for analysis using the Partial Least Squares Structural Equation Modeling (PLS-SEM) method in SmartPLS 4, namely the measurement model and structural model tests. The analysis results show that service quality is not significantly related to user satisfaction but has a direct effect on trust. Meanwhile, information quality, system quality, and human-like chatbot interaction are positively related to trust and user satisfaction. Trust and user satisfaction are positively related to intention to continue using. The results of this study can contribute both theoretically and practically to modern AI chatbot quality evaluation models.

Client-side brute force attack detection leveraging large language models

Israa A. Abbas, Taha M. Hasan, Ruaa A. Suhail — Mon, 20 Jul 2026 00:00:00 +0000

For email authentication systems, the last mile of authentication poses a critical security challenge, as attackers leverage known, real logins to bypass established server-side protections against brute-force attacks and credential stuffing. Traditional methods like rate limiting and threshold anomaly detection are ineffective against distributed, stealthy attacks that mimic normal user activity. This paper proposes a novel framework for detecting authentication sequences on the client side by leveraging lightweight Large Language Models (LLMs) in real time. Unlike signature-based approaches, the proposed framework can "explain" the temporal and contextual characteristics of login behavior and detect anomalous sequences, such as high-frequency bursts and low-and-slow last-mile reassembly attacks, without the need for predefined signatures or centralized log aggregation. The framework was tested with 2.5 million authentication sessions, 1 million of which were marked as attacks, and achieved 98.4% attack-detection accuracy with a false-positive rate of 1.2%. With an average response time of 180 ms, it achieves near real-time inference without impacting the user experience. The system was live-deployed for six months and succeeded in blocking more than 1.2 million malicious logon attempts and safeguarding more than 850,000 user accounts against traditional baselines in both scale and precision. The results demonstrate the promise of lightweight, client-side LLMs as a privacy-safe, low-resource, and robust solution for last-mile email authentication security against current and future brute force attack vectors.

Hybrid CNN-LSTM model for urban air pollution forecasting: towards early warning systems for public health: a case study of Baghdad and Basra

Israa Nadheer, Mohammed Imad Aal-nouman — Tue, 21 Jul 2026 00:00:00 +0000

Air pollution has become a critical issue in environmentally challenged regions due to rapid urbanization, especially in countries where forecasting tools remain underdeveloped. Despite the success of deep learning approaches in predicting air quality, there has been little research on multivariate forecasting of particulate matter in Iraqi cities using publicly accessible environmental information. This study proposes a CNN-LSTM architecture to predict the concentrations of PM₂.₅ and PM₁₀ the following day in Baghdad and Basra. Environmental data related to air quality, along with temperature, relative humidity, wind speed, and seasonal cyclical features, were used to generate multivariate time series with a 14-day lookback window. The proposed model was compared with an LSTM model and a conventional autoregressive integrated moving average model using 1,249 daily data samples collected from August 2022 to December 2025. The results reveal that the deep learning models outperformed the statistical baseline, and the LSTM approach exhibited the best generalization performance. A threshold-based pollution alert module was also implemented to demonstrate how continuous concentration forecasts can support operational air quality warning systems. The findings demonstrate the applicability of multivariate deep learning for short-term urban air pollution forecasting in data-constrained environments and provide a baseline framework for future environmental monitoring systems in Iraq.

Deep Q-network path planning with stratified sampling and prioritized replay

Yuxing Zhang, Setyawan Widyarto — Tue, 21 Jul 2026 00:00:00 +0000

Deep reinforcement learning (DRL) has been widely adopted for mobile robot path planning in unknown and complex environments. However, existing DRL-based approaches suffer from two main limitations: sparse reward, which provides meaningful feedback only at terminal states, and experience heterogeneity, which reduces the efficiency of uniform sampling from the standard experience replay buffer due to the diversity of transition qualities stored within it. To tackle both issues at once, we propose SS-PER-DQN, a novel path-planning approach in which stratified sampling and a prioritized experience replay mechanism are incorporated into the Double Deep Q-Network. The memory buffer is divided into three priority layers based on temporal-difference error levels, with a dynamically varying sampling ratio between layers, and importance-sampling compensation to balance the sampling-distribution bias. A densely shaped reward function is additionally employed to expedite early exploration in the presence of sparsely rewarded actions in the environment. Experiments performed on Grid World and ROS Gazebo simulations show that the SS-PER-DQN is able to successfully plan paths at success rates of 94.5% and 89.0%, with a 24.5% and 21.8% reduction in path lengths, respectively, when compared to the traditional DQN algorithm, after convergence in approximately 53% fewer training episodes. The analysis conducted during the ablation study reveals the specific roles of each contribution: stratified sampling helps the algorithm converge early by ensuring that informative state transitions guide the first phase of learning, and importance weighting influences the quality of gradient updates in later stages. It is also observed that the absolute planning time per decision step ranges from 5.0 ms (Grid World) to 5.3 ms (Gazebo) for SS-PER-DQN, representing an incremental overhead of approximately 1.9–2.2 ms per step relative to the classic DQN baseline, which remains well within the real-time requirements for robot navigation.

Techno-economic assessment of commercial wind turbines for electricity generation in Nigeria

Youssef Kassem, Hüseyin Çamur, Muhammad Mahdi Eniola Lasisi — Wed, 22 Jul 2026 00:00:00 +0000

The main aim of the present paper is to examine the wind energy potential, as well as the performance and economic sustainability, based on mean monthly wind speed data collected from the Nigerian Meteorological Agency (NIMET) and previous scientific studies. In this study, the Weibull distribution function is used to analyze the characteristics of the wind resources of 32 locations in Nigeria. Besides, intensity of turbulence, wind power density (WPD), annual energy production (AEP), capacity factor (CF), cost of energy (COE), and payback period (PBP) were evaluated for six commercial Enercon wind turbines (E53/800, E82/2000, E70/2300, E82/2300, E82/3000, and E101/3000). Results showed considerable differences between seasons and across space. Jos and Obudu were found to possess significant wind energy potential, with mean wind speeds above 8.0 m/s and WPD greater than 300 W/m². In contrast, Delta, Edo, Bauchi, and Abia had very low wind energy potentials, having WPDs less than 10 W/m². The seasonal analysis indicated that about 80% of the selected regions had high wind energy potential during the rainy season, resulting in greater energy generation capacity and improved turbine performance. In terms of energy production and cost per unit generated, the Enercon E101/3000 was shown to perform better than other turbine types, generating the greatest annual energy while recording the lowest cost per unit. The minimum energy cost of about 0.014 USD kWh⁻¹ and the shortest payback period of 1.40 years were observed in Jos, while the maximum energy cost per unit and payback period were recorded in Bauchi. Based on these results, Jos and Obudu appear to be the best candidates for installing large-scale wind farms in Nigeria.

Leakage-aware federated-style simulation for fraud risk prediction in AI-enabled FinTech platforms

Jyoti Agarwal, Moin Uddin — Sat, 25 Jul 2026 00:00:00 +0000

AI-enabled FinTech platforms increasingly rely on automated fraud-risk prediction to secure high-volume digital payments, mobile-money transactions, and real-time financial services. However, fraud detection remains difficult because fraudulent transactions are rare, financial data are sensitive, and centralized machine-learning models may create privacy and data-governance risks. This study evaluates a leakage-aware and imbalance-aware federated-style simulation for fraud-risk prediction in AI-enabled FinTech platforms. The study does not implement FedAvg, cryptographic secure aggregation, differential privacy, or model-parameter exchange. Instead, the PaySim Synthetic Financial Dataset was partitioned into five simulated FinTech clients, local models were trained separately, and prediction-level aggregation was used to compare centralized and simulated decentralized learning settings. The experimental workflow included data-quality assessment, leakage-aware feature selection, feature engineering, class-imbalance analysis, centralized model development, threshold optimization, and federated-style model comparison. The results show that fraud cases represented only 0.0466% of all transactions and occurred only in CASH_OUT and TRANSFER operations. The best-balanced centralized model was the Improved Random Forest at a threshold of 0.33, achieving an F1-score of 0.3019. Federated Ensemble Aggregation achieved the highest PR-AUC of 0.2699 and a precision of 1.0000, but its recall remained low at 0.1429. The centralized Improved Random Forest retained the highest F1-score of 0.3019, indicating a stronger overall precision-recall balance. Global feature-importance analysis showed that transaction amount, log-transformed amount, temporal features, transaction type, and destination-account activity were the main fraud-risk indicators. These findings should be interpreted as evidence from simulated decentralized fraud-risk learning rather than a full secure federated learning implementation.

A hybrid heuristic–reinforcement learning method for multi-table join optimization in cloud-native databases

Huichen Ma, Shaochun Liu — Mon, 27 Jul 2026 00:00:00 +0000

Multi-table join optimization in cloud-native databases faces challenges of storage-compute separation, resource elasticity, and runtime cost variation that classic optimizers cannot handle, which drives the need for methods able to adjust their decision boundaries according to the runtime behavior of the underlying cloud infrastructure. The proposed hybrid architecture freezes the stable sub-components of an IKKBZ-GOO heuristic seed plan into a structural freeze mask that constrains the learning algorithm throughout training, rather than being discarded after a single warm-start iteration. The mask works together with a state-aware cost function whose single state-dependent parameter is updated online. Two further components complete the design: a logistic gate trained on production telemetry, and a regression containment strategy that falls back to the heuristic plan when the learned plan underperforms. Training converges in 6.4 hours on one 16-vCPU compute node without GPU acceleration. The method was evaluated on the Join Order Benchmark, TPC-H, and TPC-DS, with PostgreSQL running on an eight-node Kubernetes cluster with object-store-backed shared storage. It achieves a geometric mean speedup of 1.38-fold on JOB, 1.28-fold on TPC-DS, and 1.07-fold on TPC-H, produces a query plan in 4.6 to 6.4 ms, returns median latency to within 5% of its pre-scaling value within 22 seconds of a scaling event, and incurs a 21% overhead at the 95th percentile under node jitter. These findings demonstrate that heuristic and machine-learning components can be integrated to yield a usable optimizer whose decision boundaries adapt to the environment in which it is deployed.

Advances in computational fluid dynamics for enhanced heat transfer applications in engineering systems: a critical review and comparative synthesis

Mon, 17 Aug 2026 00:00:00 +0000

Computational Fluid Dynamics has become an important tool for evaluating heat-transfer enhancement in engineering systems where thermal performance must be balanced against pressure loss, energy consumption, and design constraints. This critical review and comparative synthesis examine published CFD investigations of selected heat-exchange applications, including shell-and-tube, double-pipe, twisted-tube, and ground-air heat exchangers. The reviewed studies include nanofluid-based enhancement, geometric modification, internal turbulators, and optimization-based design strategies. The selected literature is compared in terms of evidence for heat-transfer improvement, hydraulic penalties, numerical modeling options, validation methods, and practical applications. The synthesis shows that geometric changes and passive flow-disruption methods can yield significant thermal gains, but at the cost of higher-pressure losses. Nanofluids are also a potential means of further improving heat transfer with relatively minor structural modification, but their long-term stability, fouling characteristics, and pumping needs are still practical issues to be addressed. The comparison also shows that no single enhancement method is always best; performance depends on the exchanger configuration, operating conditions, flow regime, and design objectives. The review highlights the need for consistent thermal–hydraulic performance assessment, better experimental validation, and more attention to manufacturability and long-term operation. It concludes with application-oriented guidance and future priorities for reliable, energy-efficient CFD-assisted thermal-system design.

Wireless sensor network-based real-time monitoring system for large ship airbag launching: design and field validation

Tue, 18 Aug 2026 00:00:00 +0000

Traditional monitoring solutions for airbag launching of large ships rely on mechanical pressure gauges with limited spatial coverage, slow response times (from several hundred milliseconds to a few seconds), and an inability to monitor dynamically changing attitude parameters during the launching procedure. This paper introduces a WSN-based real-time monitoring solution that combines IMUs and pressure sensors installed at strategic positions on the ship hull and airbags. The solution uses the hierarchical Zigbee mesh with edge computing nodes and the TDMA protocol to deliver real-time, high-speed communication with low end-to-end latency, even under challenging shipyard conditions. The Kalman filter fuses multi-sensor data to provide comprehensive monitoring of pitch, roll, heave displacement, velocity, and acceleration. In-field validation tests have been performed on cargo ships with 1500-2000 DWT capacities during the actual launching process at cooperating shipyards. The system achieves a sampling rate of 100 Hz for inertial data and 50 Hz for pressure data, end-to-end latency under 60 ms (95%), and measurement accuracy of ±0.3° for attitude angles and ±1.5% full scale for pressure measurement. Comparative analysis showed that the proposed system outperforms traditional mechanical gauges with a 75% reduction in spatial blind spots (60%→15% hull coverage), 20–100× faster temporal resolution (50–100 Hz vs. 1–5 Hz), and an 85% reduction in installation time (3.0 h vs. 19.5 h).

A TF-IDF and logistic regression baseline for binary sentiment classification of short texts

Mustafa Ali Mohsin, Salam Abdulkhaleq Noaman — Mon, 31 Aug 2026 00:00:00 +0000

The rapid growth of social media has created a large amount of short-text data that expresses people's feelings and opinions on many topics; therefore, sentiment classification has become a popular problem to address. This paper focuses on binary sentiment classification using the Sentiment 140 dataset of 1.6 million labeled tweets. A preprocessing pipeline, which includes text normalization, URL removal, and noise cleaning, was applied to the original dataset. Text representation was conducted using term frequency-inverse document frequency with an n-gram model. Logistic Regression algorithm was used for classification. The paper includes experimental evaluations using common metrics, such as accuracy, precision, recall, F1-score, ROC-AUC, the confusion matrix, and the precision-recall graph. Results show that the combination of TF-IDF feature extraction and a Logistic Regression classifier produced consistent results and achieved a computationally inexpensive baseline classifier with 85.2% accuracy, 0.852 F1-score, and 0.93 ROC-AUC when performing the Sentiment140 binary classification task. These results strongly suggest that traditional machine learning algorithms can still provide highly accurate models for such a large-scale classification challenge. Despite this, the current model has limitations in understanding sarcasm and linguistic idioms, suggesting an opportunity for deep learning and transformer-based models to outperform.

A fragile reversible watermarking technique for medical imaging security using prediction error expansion and Canny edge detection

Asaad F. Qasim, Ahmed Hussain Ali — Tue, 01 Sep 2026 00:00:00 +0000

This paper develops a fragile, reversible watermarking technique to detect accidental and intentional alterations in DICOM medical images. A fragile method is adopted to detect any manipulation applied to medical images, while reversibility ensures the recovery of the original unaltered image at extraction. The proposed approach automatically segments medical images into Region of Interest (ROI) and Region of Non-Interest (RONI), extracts authentication data from the DICOM header, and generates a digital signature of the image. The method combines integrity and authentication data and inserts them into the edge areas within the ROI, implementing the Prediction Error Expansion (PEE) technique and Canny edge detection. During extraction, the embedded watermark and the original unmanipulated image are restored without knowing the pixel locations used to hide the watermark data. The method was assessed using twelve 16-bit DICOM images, including MRI and CT, with a 512x512 pixel size. The experimental results yielded an average PSNR of 97.19 dB and an average SSIM of 0.999996, with an average watermark length of 7835 bits per image. The proposed method also fully met the reversibility requirement by recovering both the original unaltered images and the embedded watermark data. Fragility evaluation showed that manipulations of watermarked images can be detected via alterations in the retrieved watermark. The achieved results demonstrate the effectiveness of the proposed technique in maintaining visual image quality while ensuring reversibility, integrity, and authenticity in the conducted experiments.

A transformer-based explainable machine learning framework for heterogeneous ESG risk and return prediction using text intelligence

Feifei Fan, Asri Marsidi — Thu, 03 Sep 2026 00:00:00 +0000

Corporate ESG disclosure has shifted toward narrative form, yet vendor ESG ratings that most empirical work depends on show substantial disagreement across providers, and encoders dominant in financial text analysis inherit a 512-token ceiling that discards most of a typical 10-K. HET-ESGFormer combines hierarchical long-document encoding, heterogeneity-aware sparse expert routing conditioned on industry sector and firm size, and a dual-channel explanation layer subject to faithfulness verification. The framework is evaluated on 3,182 U.S.-listed firms over 2015 to 2024 using 10-K filings, sustainability reports, earnings call transcripts, and ESG-tagged news. Against nine baselines spanning econometric, dictionary-based, contextual encoder, sparse-attention, and time-series approaches, HET-ESGFormer reduces return RMSE from 0.0417 to 0.0389 and QLIKE from 0.148 to 0.128, with Diebold-Mariano tests rejecting equal predictive accuracy against every baseline after correction for multiple comparisons. The text channel's marginal contribution to downside risk prediction is materially larger than its contribution to return prediction, widening at longer forecast horizons, indicating that ESG narratives carry more information about risk than about expected return. Ablations reveal a decoupling between predictive skill and explanation faithfulness: removing the consistency regularizer barely affects accuracy but sharply degrades explanation quality. The dual-channel explanation improves ERASER comprehensiveness by 50% over SHAP applied to XGBoost, and human raters significantly prefer the framework's rationales over SHAP-based rationales on trustworthiness. Decile-sorted long-short portfolios deliver a post-cost Sharpe ratio of 1.31 against 0.74 and 0.62 for portfolios built on Refinitiv and MSCI ratings, and pass VaR backtests at 95% coverage. The results support text-based ESG modeling as an alternative to vendor-rating-based approaches, particularly where downside risk management and traceable explanation are jointly required.

An intelligent distance-driven community detection framework for complex networks using the longest distance node head technique (LDNHT)

Srinivas Amedapu, Leela Velusamy R — Fri, 11 Sep 2026 00:00:00 +0000

Community detection in complex networks needs methods that are not only structurally good, but also computationally efficient, stable, and interpretable. This study introduces a centroid-based community detection framework called Longest Distance Node Head Technique (LDNHT), where the first centroid is chosen randomly, and the others are determined by maximizing the total shortest-path distances from existing centroids. The method was tested on Facebook, Power-Grid, Wiki-Vote, and Gnutella networks in comparison with Random Node Head Technique (RNHT), Highest Degree Node Head Technique (HDNHT), classical max-min distance seeding, Louvain, and Fast-Greedy community detection. The stochastic methods were tested on more than 35 seeded runs, and the deterministic baselines were performed under the same experimental protocol. The modularity, External Connection Ratio (ECR), conductance, coverage, and runtime were used to assess partition quality. On Power-Grid and Wiki-Vote, its performance was significantly better than RNHT, HDNHT, and MAXMIN in terms of ECR and coverage, while Louvain and Fast-Greedy generally achieved higher modularity. Sensitivity analysis showed that the number of repeated initializations had a small impact, with the dependence on the first random centroid decreasing significantly with repeated initializations, and the benefit diminishing after about 20 internal iterations. The empirical scalability experiments on sparse networks showed approximately linear behavior, with a log-log slope of 0.911. Overall, LDNHT offers a comprehensible cumulative geodesic-dispersion strategy whose effectiveness is dependent on network structure and parameter configuration.

COVID-19 diagnosis using a decision tree algorithm

Zahoor M.Aydam, Tasadi M. Hanoon, Ghosoon K. Munahy — Sat, 12 Sep 2026 00:00:00 +0000

Coronavirus disease 2019 (COVID-19) is a respiratory infection for which chest X-ray (CXR) imaging can provide an accessible screening modality. This study investigates an interpretable computer-aided screening pipeline for distinguishing COVID-19-positive from normal CXR images. Methods: The pipeline combines Otsu-based lung segmentation, morphological refinement, seven Hu invariant shape moments, training-fold z-score normalization, and a class-weighted Decision Tree. A cohort of 2,600 CXR images (1,000 COVID-19-positive and 1,600 normal) was evaluated using stratified five-fold cross-validation. The retained experimental record reports 98.6% accuracy, 97.8% precision, 98.6% recall (sensitivity), and 98.2% F1-score. The pooled confusion matrix shows TP=986, FN=14, FP=22, and TN=1578, corresponding to 98.6% accuracy and 97.8% precision. ROC/AUC analysis yielded AUC = 0.987 ± 0.004 under the image-level evaluation protocol. Conclusion: The findings indicate that a low-dimensional, shape-based descriptor combined with an interpretable Decision Tree can discriminate COVID-19-positive from normal CXR images under the present internal image-level protocol. Because the cohort was not partitioned at the patient or source level and no independent external cohort was retained in the study record, the results are considered preliminary proof-of-concept evidence rather than clinically validated diagnostic performance.

Global value chains, domestic value creation, and firm performance: machine learning evidence from Vietnam

Huynh Thi Thuy Giang, Nguyen Anh Phong, Tran Huong Thi — Mon, 14 Sep 2026 00:00:00 +0000

This study examines the impact of domestic value added and participation in global value chains (GVCs) on firm performance, based on a sample of 287 observations from listed companies across five key industries in Vietnam's Southeast region during 2020-2024. Three machine learning methods, Ridge Regression, Random Forest, and Gradient Boosting, are applied to estimate the influence of each independent variable on firm performance. Ridge regression produced the best estimates of firm performance among these three model types, based on the lowest RMSE (4.97), MAE (2.78), and R-squared (0.57). This suggests that the relationship between GVC-related attributes and firm performance is largely linear and additive. Among the individual predictors, lagged profitability was the most important single predictor of current performance, followed by GVC activity. Notably, domestic added-value predicted firm performance more strongly than foreign added-value. Greater GVC activity therefore does not necessarily translate into stronger firm performance. Instead, firms appear to perform better when global engagement is combined with strong internal capabilities and substantial domestic value creation. Firm size, financial leverage, technological capability, and logistics efficiency also statistically explained variation in firm performance.

AI-driven multi-objective optimization of CNC machining parameters for enhanced surface finish and tool life

Wed, 16 Sep 2026 00:00:00 +0000

AI-driven CNC machining optimization requires balancing surface quality and productivity while avoiding unsupported integration of independently collected machining data. This study proposes a leakage-controlled, surrogate-assisted framework using two open datasets: a 27-condition turning surface-quality dataset and a 968-observation milling tool-life dataset comprising 14 tools. The datasets were analyzed independently, with no row-wise fusion or DOC–ADOC mapping. Cutting speed, feed rate, and depth of cut were used as controllable turning variables, while milling sensor and metadata features were used separately for remaining useful life (RUL) prediction. The selected Ridge surface model achieved LOOCV MAE = 0.1077, RMSE = 0.1383, and , while nested validation yielded , indicating limited out-of-sample generalization. The no-cycle RUL model achieved MAE = 23.44 cycles, RMSE = 31.39 cycles, and under leave-one-tool-out validation. NSGA-II significantly outperformed equal-budget random search in normalized hypervolume (0.7711 vs. 0.7421). Because the surrogate surface was statistically limited, continuous optimization was treated as exploratory. Final selection was therefore anchored to measured Pareto-efficient turning conditions. Equal-weight TOPSIS selected , , and , with measured and , as the balanced practical recommendation.

XLeaf-Net: an explainable deep learning framework for mango leaf disease classification and severity assessment using Grad-CAM-based visual interpretation

Shweta Amar Satao, Swati R. Maurya — Fri, 18 Sep 2026 00:00:00 +0000

Accurate mango leaf disease recognition is important for automated crop-health assessment, but reliable evaluation requires transparent data partitioning, appropriate baseline comparison, and cautious interpretation of severity and explainability outputs. This study presents XLeaf-Net, an explainable deep-learning framework for mango leaf condition classification, image-derived severity assessment, and Grad-CAM-based visual interpretation. Experiments were conducted on the balanced MangoLeafBD dataset using an image-level stratified train-validation-test protocol with file-level integrity and capture/session-group overlap checks. XLeaf-Net was trained from scratch and compared with from-scratch VGG16, ResNet50, and EfficientNetB0 models under the implemented model-specific training configurations. An auxiliary severity model used conservative, image-derived symptom-intensity proxy labels, while Grad-CAM was applied to qualitatively examine class-discriminative regions. XLeaf-Net achieved 98.67% image-level test accuracy and a 98.66% macro F1-score, outperforming the evaluated from-scratch baselines while using 2.52 million parameters. The severity model achieved 78.17% accuracy and an 80.89% macro F1-score. Grad-CAM visualizations frequently overlapped with visually symptomatic leaf regions in representative correctly classified images. These findings demonstrate strong image-level performance within the implemented protocol; however, severity labels were not expert-annotated, and the evaluation was not session-independent. XLeaf-Net should therefore be considered a compact candidate framework requiring further independent validation.

Probabilistic-temporal modeling of network security monitoring subprocesses for multi-stage attacks in IoT networks

Ammar Dawood Jasim, Mokhalad Al-Tameemi — Sat, 19 Sep 2026 00:00:00 +0000

Timely detection of a multistage cyberattack in an IoT network depends on the chosen monitoring architecture, execution time, and the transition behavior of subprocesses within that architecture. This paper introduces a probabilistic temporal model to evaluate an NSMS for both known and unknown multi-stage attacks. First, we introduce a conceptual monitoring architecture that organizes streaming-data processing, temporal analysis, attack classification, attack prediction, and response selection. The recurrent neural networks in the architecture form the basis of the implementation framework and are neither trained nor empirically analyzed. The evaluated contribution represents the NSMS subprocesses as a continuous-time Markov chain, where each state corresponds to a monitoring task and each transition rate is estimated from the respective mean processing time. We develop state-probability equations and use a specialized software tool to estimate the stationary probabilities and successful anomaly-detection probabilities. Numerical evaluations examine how delays in preprocessing, syntactic construction, encoding, classification, and prediction affect monitoring effectiveness. The analysis shows that the estimated probability of successful anomaly detection decreases as preprocessing and encoding time increases, underscoring the need to account for computational delay in time-critical security monitoring. The sensitivity and steady-state analyses identify the subprocesses that most affect the modeled system's behavior. The developed framework enables quantitative evaluation of NSMS timing constraints and allocation of computational resources before designing and experimentally evaluating the learning-based architecture.

Spatial–temporal machine learning for traffic violation type prediction: a case study in Qatar

Mohammed Alshriem, Yin Yang — Mon, 21 Sep 2026 00:00:00 +0000

Although traffic violations are among the most significant factors contributing to road accidents, little research has focused on predicting specific types of traffic violations in Qatar and the Gulf region, where the number of vehicles has increased significantly in recent years. This work addresses this gap by proposing a machine-learning framework to solve fine-grained, multi-class prediction of traffic violation types on a large-scale dataset from Qatar. The dataset is extremely imbalanced and has a high-dimensional categorical space, with 97 original violation classes. We created a preprocessing pipeline to address these issues by filtering the 97 original codes to nine well-represented violation types while retaining 99.6% of the records. The final model uses 18 spatiotemporal features, including our zone violation entropy measure, which captures violation diversity in each enforcement zone. Five classifiers are evaluated with a strict temporal train–validation–test split, training-based encoding (to avoid temporal leakage), and balanced class weighting. Random forest's macro F1 score is 0.6184 and balanced accuracy is 0.7064, outperforming the linear and neural baselines. SHAP analysis shows that location-derived features account for ~74% of the model's predictive importance, while temporal features account for ~25%. The results indicate that violation type is more strongly associated with location than with time, and they provide implications for targeted enforcement by location. The suggested framework provides a reproducible way to predict violation types in a multi-class setting. It shows that enforcement information alone can support predictive analysis without using road topology or sensor data.

Cybercrime law and digital forensics: adapting legal frameworks for future technologies

Mon, 21 Sep 2026 00:00:00 +0000

Linked authenticity, integrity, reliability, attribution, jurisdiction, and privacy challenges are created by digital evidence from cloud platforms, Internet of Things (IoT) devices, artificial intelligence (AI), large language models (LLMs), biometrics, drones, blockchain, and virtual environments. This review synthesizes scholarly and authoritative sources on cybercrime law, digital forensics, and emerging technologies through a structured narrative review. The review reveals a technical–legal divide: it emphasizes investigative efficiency, while acquisition, hashing, metadata, chain of custody, tool validation, and automated-analysis reliability require clarification for evidentiary scrutiny. Technology, evidence source, forensic challenge, legal risk, admissibility requirement, and governance response are linked through the Technology–Forensic–Legal Admissibility (TFLA) Framework. Primary artifacts are distinguished from analytical outputs; technical weaknesses are translated into legal consequences and safeguards. Standards and evidentiary rules are complemented; TFLA is conceptual, not empirically validated. Technology-neutral, technically informed, rights-compliant governance is supported through transparent documentation, corroboration, human oversight, and institutional competence. Digital-forensic system design is linked to trustworthy technology governance.

Deep transfer learning and radiomics synergy for enhanced lung cancer classification on CT scans

Mohammed S.H. Al-Tamimi, Amenah H. Abdulateef, Isra H. Abdulateef — Thu, 24 Sep 2026 00:00:00 +0000

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, making accurate and timely detection essential for improving patient outcomes. Computed tomography (CT) is widely used for lung assessment; however, manual interpretation can be affected by inter-observer variability and subtle imaging characteristics. This study proposes an automated three-class lung CT classification framework that integrates deep transfer learning with quantitative lung-region features. EfficientNet-B7, initialized with ImageNet-pretrained weights, was employed to extract high-level deep representations from CT images. In parallel, lung Regions of Interest (ROIs) were automatically segmented, from which four quantitative descriptors were extracted: Skewness, Kurtosis, GLCM Contrast, and GLCM Homogeneity. The standardized lung-ROI features were subsequently concatenated with the EfficientNet-B7 deep feature representation using a feature-level fusion strategy and classified into Normal, Benign, and Malignant categories. Experimental evaluation on the held-out test set demonstrated an overall classification accuracy of 96.36%, correctly classifying 106 of 110 CT images. The proposed fusion framework achieved macro-averaged Precision, Recall, F1-score, and Specificity of 94.39%, 90.87%, 92.39%, and 97.53%, respectively. Furthermore, one-vs-rest ROC analysis yielded AUC values of 0.9849, 0.9634, and 1.0000 for the Normal, Benign, and Malignant classes, respectively, corresponding to a macro-average AUC of 0.9828. These results demonstrate the strong discriminative capability of the proposed deep and quantitative feature fusion framework for three-class lung CT image classification.

A new hash function based on irreversible finite automata

Ghassan Salloom, Layth Hassnawi, Karam Mohammed, Mohammed Adhab — Thu, 24 Sep 2026 00:00:00 +0000

A mathematical hash function maps inputs of arbitrary length to unique outputs (digests) of a fixed length. It is highly useful and used in almost all information security applications; it can also serve as index data in hash tables, detect duplicate data for fingerprinting or uniquely identifying files, and serve as well for checksums to identify data corruption. In this work, a novel 256-bit cryptographic hash function based on the irreversible Mealy finite automata model (IFA) is introduced. In this architecture, the irreversible transition matrix of the Mealy automaton (IFA) is constructed using a non-injective state-transition mechanism to introduce local irreversibility and ambiguous backward reconstruction. The automaton output table is precomputed using a Chebyshev-based nonlinear generator to improve output randomness and nonlinear complexity. For further diffusion, confusion, collision resistance, and one-wayness improvements, we integrate the irreversible automaton within a sponge-based construction with a 512-bit internal state (256-bit rate + 256-bit capacity), yielding a designed collision resistance of bits and preimage resistance of under the sponge capacity bound. Evaluation against SHA-256 shows an avalanche mean of 50.02% (σ = 3.14%), all 13 NIST SP 800-22 statistical tests passed, and zero collisions among 50,000 tests. Throughput is about 13 MB/s versus 183 MB/s for SHA-256, reflecting the sequential dependency of the Mealy transformation.

A deep learning-driven adaptive edge intelligence framework for real-time Internet of Things applications

Majd S. Ahmed — Fri, 09 Oct 2026 00:00:00 +0000

The rise of Internet of Things (IoT) ecosystems and the New Paradigm of low-latency edge data processing. IoT ecosystems are growing exponentially, and this growth brings a new challenge: low-latency data processing at the network edge. Because cloud-based architectures cannot meet this real-time requirement due to network delay, connectivity issues, and increasing bandwidth contention, deploying deep learning-based, high-computational-load pattern detection on high-dimensional streaming IoT data becomes challenging. The proposed framework adapts the model's computational complexity and inference location in response to system state to optimize latency and energy while retaining high predictive accuracy. Experimental evaluation on the NSL-KDD intrusion detection dataset shows high accuracy, with class-wise accuracy of 0.979, precision of 0.981, recall of 0.977, F1-score of 0.979, and a precise ROC area under the curve of 0.998. Comparisons with simulation results provide useful insights into the relative advantages of local edge inference over end-to-end cloud inference, yielding 98% energy savings and a 97.6% latency reduction. This work reports a novel context-aware integrated framework that combines a lightweight deep learning classifier, adaptive task offloading using a Deep Q-Learning-based intelligent agent, hierarchical workload distribution at edge-fog-cloud levels, and proactive contextual information dissemination in real time, in order to ensure a context-aware system with 58.5%, 29.0%, and 12.5% edge, fog, and cloud-based decisions, respectively.

H-VSAF: a verifiable and privacy-preserving federated learning framework for secure remote healthcare monitoring

Sureshkumar S, Santhoshkumar S.P, Joseph James S, Priya R — Sat, 10 Oct 2026 00:00:00 +0000

The sensitive patient data handled by remote healthcare monitoring systems requires reliable, secure learning methods. Federated learning enables collaborative healthcare analytics without storing raw medical data centrally. However, existing techniques still suffer from issues including privacy leakage, untrusted aggregation, and frequent device outages. In this work, we propose H-VSAF, a privacy-preserving, verifiable, and lightweight federated learning system for healthcare environments. H-VSAF uses linear homomorphic hashing, Bloom filter-based online authentication, Shamir secret sharing, Diffie-Hellman key generation, and single-mask Secure Aggregation to ensure secrecy, precision, and resilience. The security analysis provides informal guarantees against gradient inference, aggregation tampering and deletion, and forged dropout claims under the threat model and cryptographic assumptions above. We experimentally evaluate H-VSAF on healthcare datasets and demonstrate that it achieves model accuracy within 0.2-0.8% of standard federated learning while decreasing client compute cost by up to 32% and communication overhead by 28% compared to traditional secure aggregation methods. Moreover, the framework is resilient to 40% client dropouts and introduces verification latency < 35 ms each round. The results clearly show that H-VSAF provides a credible and practical platform for safe joint health care analytics.

Assessment of somatic excitability and dendritic responses in a reduced conductance-based CA1 neuron model

Youwang Yao — Mon, 12 Oct 2026 00:00:00 +0000

A reduced conductance-based model was assessed through somatic excitability, firing dynamics, and dendritic conductance responses. The reported baseline spike had a threshold of −36.5 mV, a baseline-to-peak amplitude of 115.1 mV, and a half-width of 0.93 ms. At 8 µA/cm², ISIs shortened before lengthening by 17.8% relative to the minimum interval. Dendritic A-current blockade increased amplitude from 92.8 to 100.8 mV, and halving dendritic Na⁺ conductance reduced it to 69.9 mV. Two new numerical experiments used a complete reference implementation with published channel kinetics. Joint waveform retuning reduced the normalized feature error by 84.7%, shifting the threshold from -39.13 to -49.36 mV and the amplitude from 114.71 to 96.76 mV. Matched zero-M and increased-M controls linked M-gate availability to late ISI lengthening, while a fixed A-inactivation gate changed the early-interval trajectory. The reported and reference baselines use their respective specified kinetics. These analyses quantify waveform trade-offs and conductance-dependent firing in the compact representation.

PoET²S: Folding a digital twin into validator selection for blockchain consensus in peer-to-peer energy trading

Tue, 20 Oct 2026 00:00:00 +0000

Blockchain platforms for peer-to-peer energy trading often add a digital twin that forecasts each meter's generation and demand. In most designs the twin only raises alarms, while validators are still chosen by stake or computation. The proposed scheme, PoET²S, incorporates the twin's evidence into validator selection: a node's weight is a stake-and-energy term multiplied by its Twin-Trust Score raised to γ = 2.5, so nodes that contradict the physical forecast quickly lose committee seats. A discrete-event simulator compares PoET²S (500 nodes, 15% Byzantine, 30 paired seeds) with seven classical baselines and three design points from recent work: a binary veto, an additive energy-and-trust score, and a linear (γ = 1) reputation weight. Detection is a near-tie: PoET²S catches 91.0% of attackers and the veto 91.7%. PoET²S raises more false alarms per catch by default (13.9 against 9.7) but matches the veto's catch rate at equal false-alarm cost. The malicious committee share is 10.8% under the veto, 3.8% under the linear weight, and 2.6% under PoET²S, and on 100 seeds the gain over the linear weight is 37% (Holm-adjusted p = 7.2 × 10⁻⁹). At 45% Byzantine nodes, the one-third fault bound fails in 64.4% of veto blocks and 2.4% of PoET²S blocks. Simulated commit latency (30.2 ms) is higher than for Proof of Stake and Delegated Proof of Stake, lower than for Practical Byzantine Fault Tolerance, and flat up to 4,000 nodes; an optional fast path lowers it to 26.8 ms. Against Sybil attacks, top-K selection needs certified meter identities.

Multi-objective reinforcement-learning simulation for co-optimizing resource circularity and environmental performance in circular manufacturing

Ting Li, Logaiswari Indiran — Tue, 27 Oct 2026 00:00:00 +0000

More regulations require manufacturers to enhance material circularity and reduce their environmental footprint simultaneously; however, it remains unclear whether these two objectives align, and if so, in which domain. A discrete event simulation model for a remanufacturing plant operating multiple production lines with environmental performance measures at the process level is defined as a multiobjective Markov decision process, which is then solved using a preference-conditioned actor–critic technique, employing feasibility masking for the precedence and capacity constraints and adaptive sampling between the surrogate model and the full-scale simulator. Two life cycle accounting methods have been used: Boundary A (cut-off) and Boundary B (avoided burden). The environmental burden reveals three regimes, synergistic, trade-off, and inverted, characterized by transition points in secondary-material input levels at 0.38 and 0.61 for Boundary A and at 0.47 and 0.78 for Boundary B; sweeps with fixed ratios validate these transition points. One conditionally generated network accounts for 87.4% of the non-dominated solutions of eleven scalarized agents using 9.1% of the simulation budget allocated to them, and the switching strategy reduces the number of simulation calls needed to reach the reference hypervolume by 27.7%. A screening based on the unit-cost criterion limits the joint synergistic and economically viable domain to input levels ranging from 0.21 to 0.38. All parameters are synthetic values from published ranges for a single facility, so the boundaries describe a mechanism rather than plant-specific forecasts.