Browse Items (16481 total)
Sort by:
-
A Comparative Evaluation of Standalone LLMs and Retrieval-Augmented Generation Models Using Hypothetical Gemini Systems
This study assesses the efficacy of two theoretical language models; Gemini Standalone LLM and Gemini RAG (Retrieval Augmented Generation) across diverse natural language inquiries. The assessment centers on three principal metrics: precision, pertinence, and inference duration. The experiment utilizes a controlled simulation to illustrate the benefits and drawbacks of independent language creation versus retrieval augmented generation strategies. The results demonstrate that RAG at trains superior accuracy and relevance by integrating retrieved context, albeit it incurs longer inference durations. This comparative analysis seeks to assist researchers in comprehending the ramifications of including retrieval methods into big language models. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Automated Diabetic Retinopathy Diagnosis Using Ensemble Approach
Diabetic Retinopathy is a major reason of vision impairment among diabetic patients, early and accurate diagnosis is crucial. This research focuses on developing a machine learning-based classification system to detect different stages of DR using Support Vector Machine (SVM), Random Forest (RF) and ensemble model. The dataset is divided into five categories: Healthy, Mild, Moderate, Proliferative and Severe DR. Performance evaluation using various metrics, including Accuracy, F1-score, RMSE and AUC-ROC, indicates that the ensemble model achieves the best results, with an accuracy of 77.66% and an AUC-ROC of 0.9015. The confusion matrices show that existing models struggle with certain misclassifications, the ensemble approach enhances overall predictive capability. Future improvements can include integrating deep learning models such as convolutional Neural Networks leveraging larger and more diverse datasets and incorporating image preprocessing techniques to enhance feature extraction. This system can help ophthalmologists to detect early and treatment planning, ultimately decrease the risk of blindness in diabetic patients. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Insights into the Publication Trends of Pharmaceutical Reverse Supply Chain Using Data Mining Approach
Due to the non-profitable nature of reverse supply chain of pharmaceutical products, researchers and companies have not shown much interest in this field. Due to stringent regulatory compliances pharmaceutical companies and hospitals are mandated for proper disposal of pharmaceutical wastes. This research aims to highlight the publication trends of pharmaceutical reverse supply chain using data mining approach. The metadata of published literature was extracted from Scopus and analysis was done for the title and abstracts of the articles. It was found that there is limited published literature on this topic. Co-occurrence map of text-based data, time graph of co-occurrence map of text, trigrams word cloud, keywords plus word cloud and unigrams word cloud were formed to get insights into the publication trend. A model had been proposed from the consumers end for pharmaceutical reverse supply chain. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Alcohol-Attributable Liver Disease in India, 20002021: Comparative Analysis Across Alcohol Policy Regimes Using GBD 2021
Alcohol use is ranked among the leading causes of liver disease in the world, and the most dreadful consequences of this condition are cirrhosis and hepatic cellular carcinoma (HCC). India has an eclectic policy environment, with bans, regulation, liberal paradigms, and the influence of such policies on the epidemiological process is inadequately studied. Based on the Global Burden of Disease (GBD) 2021 data of nine states (20002021), this study focuses on disability-adjusted life years (DALYs), years of life lost (YLLs), and years lived with disability (YLDs) due to alcohol-related cirrhosis and HCC. States were classified as prohibited (Bihar, Gujarat, Nagaland), regulated (Karnataka, Kerala, Tamil Nadu), and liberal (Goa, Punjab, Sikkim). Liberal states had the highest burdens, with Sikkim leading by approximately (410 per 100,000), followed by Goa (360 per 100,000) and Punjab (290 per 100,000), all above prohibited state averages. In Bihar, there was 27% reduction of DALY, whereas Kerala had the highest increase of 44%. More than 90% of total variation in DALYs was attributed to YLLs, with men also experiencing larger overall burden, ranging between 45 and 811 times during midlife. The panel regression displayed low cohort-level variance (R2= 0.41) but strong state-level effects (R2= 0.98), that signify a high level of heterogeneity. These results show that in addition to policies, variations in implementation, fiscal priorities, and social contexts determine the burden experienced in India, which further points to the need to implement evidence-supported, targeted interventions. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Bamboo Trade Dynamics: A Hybrid ARIMALSTM Forecasting Approach for Indias ExportImport Trends (20172025/26)
Bamboo plays a dual role in Indias economy, serving as both an ecological safeguard and a driver of rural livelihoods. This paper examines bamboo exportimport flows between 2017 and 2025/26 using official trade statistics. A hybrid ARIMALSTM forecasting model is implemented to capture both linear and nonlinear patterns. Results demonstrate rising exports, stable imports, and higher predictive accuracy compared to ARIMA alone, confirming bamboos growing role in sustainable trade. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
AI-Driven Fleet Management System: Smart Vehicle Directory for Organizational Efficiency
Affected by the increasing need for an effective, automated system that can manage campus entry and parking, it urgently calls for a solution to integrate license plate recognition and visitor management for security and convenience. This will further give allowances to the pedestrian visitors for a pass on their mobiles in exchange for their mobile numbers. This is a proposal for the integration of Automatic License Plate Recognition into an Automated Character Recognition system, which would help extract license plate information on a college portal. It would encompass new vehicle registrations, staff and student registrations against license plates, and would also be able to generate digital passes for visitors. Though the system depends upon the conventional algorithms, these demonstrate good performance under normal applications. This simply means that an institution is bound to ensure that campus security and parking administration run seamlessly, due to the necessity for facilitating easy and prompt vehicle registration processes and issuance of daily passes. Integration of a visitor management system issuing electronic passes via mobile numbers with an ALPR system matched with OCR for extracting license information will greatly improve the campus security in administering the parking facilities. One of the most used methodologies to this effect is You Only Look Once, with very great abilities in object detection. This will be integrated to automate license plate registration, link it to the registered users, and provide digital passes for visitors. The result will be a more organized and secure campus environment. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
A Comparative Analysis of Blockchain and Large Language Models in Enhancing Telehealth Data Security
Healthcare is rapidly moving toward utilizing technologies that enhance the delivery, management, and security of healthcare. Telehealth has greatly streamlined access to medical care, especially in managing chronic diseases. The rapid development of IoT devices and digital health records posed critical challenges in data security and privacy, among other operational issues. A centralized system is vulnerable to breaches, and therefore, there is a need for more robust and decentralized solutions. There are risks involved in the current centralized system and a new stronger decentralized system is required. The blockchain has a way of providing safe, immutable ledgers with which health data can be managed; however, its static security measures need to be improved in the context of the kind of risks and at- tacks that may appear these days. This research looks to introduce Large Language Models into blockchain-based telehealth systems. LLMs enhance blockchain security through real- time detection of threats, anomalous behavior, and optimized execution of smart contracts as LLMs contain advanced processing and pattern recognition. This system combines both the merits of blockchains decentralized architecture and the intelligent analytics of LLMs to bring a dynamic and scalable framework of telehealth security, which enhances data privacy, operational transparency, and system resilience. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Static Analysis and Machine Learning for Runtime Library Detection in Linux Binaries
The upsurge of malware targeting Internet of Things (IoT) devices demands effective approaches. This work announces a new method, stimulated by MANTILLA, which influences machine learning models. Through a prominence on architecture-independent characteristics from binary procedures, the system progresses its competence to differentiate among several libraries as well as architectures. Classification accuracy is further enhanced by employing a majority voting technique such that the output of the model is robust and reliable. Besides the machine learning-based classification, the paper incorporates a malware detection module based on signature matching. This two-pronged approach enables the system to cross-check discovered runtime libraries against a large database of pre-collected malware signatures. By marking possible security threats according to this comparison, the system greatly increases its ability to identify malicious binaries, thus offering an added layer of security for IoT devices. This unification of detection and classification mechanisms plays an important role in dealing with the changing nature of malware threats. Although encouraging results were obtained through this project, more evaluation should be done for comparison of the efficiency of KNN with other models, for example, Random Forest. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Enhanced Detection of Malicious URLs Using Supervised Machine Learning Models
This paper deliberates on URL phishing, one important subset of cyber threats. Most modern-day deceptive practices have shifted to the digital space due to the vast scope of information available on the internet. URL phishing is a dishonest practice that includes masquerading harmful links as legitimate links to trick users into sharing their private data. Detection of URL phishing is extremely challenging, hence most of these attacks go undetected until it is too late for the victim. Automatic blacklist that rely heavily on user-generated reports to monitor internet links have been repeatedly proven ineffective time and again. Along with failing to identify newly listed phishing sites, these systems also tend to mistake harmless links for phishing traps. This paper proposes the application of classification techniques of practical machine learning, specifically analysing the patterns and behaviours of URLs to detect phishing websites accurately. Leveraging the properties of Decision Trees, Random Forests, Logistic Regression, SVM, and Light GBM, we were able to come up with a detection model, which precisely calculates accuracy, precision, recall, as well as F1 score to evaluate the validity of URL classification. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Advancing Spatio-Temporal Predictive Modelling in Intelligent Transportation Systems: A Comprehensive Survey of Machine Learning and Deep Learning Approaches
In an effort to alleviate traffic and improve urban mobility, intelligent transportation systems (ITS) relies heavily on forecasting traffic. In the paper, a comprehensive survey on spatial temporal predictive modelling techniques for forecasting traffic has been presented. The focus remains on advanced machine learning and deep learning that have been developed between 2017 and 2025. With the use of state of the art technologies to forecast both in real time scenarios (short-term) traffic prediction and long term forecasting, such as transformer based models, (RNN) recurrent neural networks, convolutional networks on grids and graphs, and (GNN) graph neural networks. Former approaches were examined for strengths and limitation to capture intricate temporal dynamics and spatial interdependencies. Through the above findings, a brand-new conceptual methodology that associates attention mechanisms and graph-based learning to increase prediction accuracy with computing efficiency has been proposed. The performance improvements of newer methods over the conventional methods are also shown through a comparison of the experimental findings. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Fault-Tolerant Strategies and Federated Learning for Resilient Edge Computing: A Comprehensive Survey
Edge computing has emerged as a key technology for enabling real-time data processing in various applications such as Industrial Internet of Things (IIoT), smart cities, and autonomous systems. However, the distributed nature of edge computing makes it particularly vulnerable to system faults, such as hardware failures, network outages, and data corruption. To address these challenges, fault-tolerant strategies are essential for ensuring the reliability and resilience of edge systems. Furthermore, federated learning (FL) offers a decentralized approach to machine learning, which can enhance the resilience of edge computing environments by allowing edge devices to collaborate on training models without relying on a central server. This paper explores the integration of fault-tolerant strategies with federated learning to provide a comprehensive, resilient framework for edge computing. Various aspects are compared for fault-tolerant mechanisms with federated learning frameworks to analyze their effectiveness in enhancing system reliability and ensuring real-time performance in the face of failures. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
RL-Based Online Mutation Strategy Selection Techniques inDifferential Evolution: A Study
In recent years, reinforcement learning (RL)-based online mutation strategy selection techniques have emerged as a principled learning framework for balancing exploration and exploitation in Differential Evolution. Several state-of-the-art (SOTA) DE variants have been proposed that utilize online mutation strategy selection to improve the performance of the canonical DE algorithm. This paper presents a comprehensive review of such DE variants and studies multi-armed bandit formulations for online mutation strategy selection in DE. It systematically categorizes existing DE variants based on their utilization of RL algorithms. The Author(s) 2026. -
A Novel Hybrid Ensemble Architecture for Stroke Risk Prediction Using Healthcare Data
Stroke is the reason for an alarming number of disabilities worldwide, further emphasising the critical need for early and accurate prediction of risks to inform clinical management. This paper presents a novel hybrid ensemble architecture that leverages the superiority of multiple machine learning models for stroke health risk prediction using health data. In this novel hybridisation, decision tree classifiers belonging to the Random Forest and XGBoost families are effectively combined with support vector machines and a shallow neural network within a Stacked ensemble strategy that uses a hard vote technique. To improve model generalizability and avoid overfitting, feature selection and dimensionality reduction methods like Recursive Feature Elimination (RFE) and Principal Component Analysis (PCA) have been included expertly without compromising performance. After extensive training and testing on a real-world health repository covering a broad range of demographic, lifestyle, and clinical features, the model obtained an outstanding F1-score of 0.9427 and an exemplary ROC-AUC value of 0.9872, much higher than the performance of the individual models. Statistical significance was assessed using the Friedman and Wilcoxon signed-rank test. The model is a strong candidate for incorporation into clinical decision support systems and is fully deployable and EHR-compatible. The Author(s) 2026. -
Can LLMs Reliably Predict Personality Traits Through Prompt Engineering? Introducing Expert-Chain-Hypothesis Prompting
The prediction of personality traits from natural language is a long-standing challenge at the intersection of artificial intelligence and psychology. Traditional approaches have largely relied on handcrafted features, custom labelled datasets, supervised learning models or pretrained embeddings. However, relatively fewer studies have explored the potential of large language models (LLMs) through prompt engineering. This study investigates the capability of LLMs to reliably predict the Big Five personality traits from unstructured textual input using prompt-based methods. We introduce Expert-Chain-Hypothesis prompting, a psychology informed prompting technique that mirrors the decision making workflow of human experts within LLMs. Using the Essays dataset, we evaluate zero-shot prompting, few-shot prompting, chain-of-thought prompting and the proposed method on GPT - 4o. Experimental results demonstrate that the proposed prompting technique achieves superior overall performance; with a macro F1 score of 0.759, micro F1 score of 0.750 and the lowest reported hamming loss of 0.280. Our findings suggest that while prompt engineering techniques hold significant promise for personality prediction, challenges remain in fully capturing the subtleties of all personality traits. The Author(s) 2026. -
Decision Flow Tracing and Word Impact Analysis in Hybrid Transformer-Conditioned Diffusion Models for Text-to-Image Generation
Text-to-image diffusion models have become a cornerstone of modern generative AI, offering high-quality synthesis yet remaining constrained by their black-box nature, which limits controllability and interpretability. In this work, we propose a hybrid transformer-conditioned diffusion model that integrates UNet-based denoising with multi-head cross-attention transformer blocks at critical latent stages of the diffusion process. The architecture is trained on a curated set of 50,000 samples from DiffusionDB with a 200-step latent diffusion schedule. Text prompts are encoded using a 16-token BERT encoder and mapped into a 256-dimensional latent feature space. Cross-attention layers with eight heads are interlaced within the UNet bottleneck and decoder, enabling token-to-region correspondence and fine-grained semantic propagation. To ensure interpretability, we design an explainability framework that combines hierarchical token-level attention heat maps, temporal attention rollouts, and perceptual ablation studies based on learned image patch similarity. Analysis reveals that object tokens remain spatially and temporally consistent, while attribute tokens demonstrate sharper temporal volatility. JensenShannon divergence quantifies this redistribution of attention across diffusion steps. Experimental evaluation against a standard UNet diffusion baseline demonstrates clear improvements: Frhet Inception Distance decreases by 19.6, CLIP alignment score increases by 5.4, and Inception Score improves by 18.6. Moreover, attention coherence improves by 22%, underscoring the gains in explainability. The proposed framework establishes a pathway toward accountable, high-fidelity, and interpretable text-to-image synthesis. Beyond performance, it supports critical tasks such as bias evaluation, fairness auditing, and quality assurance, offering a robust foundation for the next generation of explainable generative AI systems. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Explainable Temporal Knowledge Graph Reasoning for Geopolitical Risk Assessment
The paper examines the application of the Graph Neural Networks (GNNs) to the ICEWS14 temporal knowledge graph to predict geopolitical events and relationship. We suggest a new architecture that involves negative sampling and temporal encoding trick that enhances performance on temporal link prediction. Our model combines a central temporal-conscious attention GNN with numerous domain-specific GNN sub-models that are taught on economics trends, political choices, and business interrelations and it additionally incorporates knowledge graphs, sentiment analysis of the people, and market tendencies data to provide comprehensive consulting assistance. The framework has an AUCROC of 0.9234, test set, indicating that the framework is more likely to detect both positive and negative links. Exploring the high-confidence predictions, we identify the regularities of the political enforcement action, whereas the low-confidence ones assist in suspecting the improbable or counterfeit connections. Such explanations are useful in designing early warning systems, risk-assessment monitors, and decision-support frameworks in foreign affairs and think tank consultancy. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Ellipse Shaped Four-Element Microstrip Patch Antenna Array Using Corporate Feed Network
The design and simulation of an ellipse-shaped four-element microstrip antenna array for effective multi-band wireless communication are presented in this paper. The proposed antenna makes use of a FR4 substrate, which has overall measurements of 240mm 150mm for a four element antenna and 55mm 55mm for a single element. Simulation of the antenna shows significant return loss at key frequencies: 0.35, 0.5, 1.7, 2.2, 3.6, 5.6, and 9.4GHz, with S11 values below ?10dB. These frequencies cover bands like radar, 4G/5G, ultra-wideband (UWB), global positioning system (GPS), and wireless LAN (WLAN). The antenna achieves broadband operation with good gain values because of the staircase-shaped ground plane and defected radiating ellipse-shaped patch. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Quantum-Driven Digital Forensics: Evidence Acquisition, Intrusion Detection, Cybercrime Simulation, andDNA Profiling
Quantum computing introduces a new paradigm in digital forensics by enabling faster cryptographic analysis, enhanced machine learning, and secure data acquisition. This research examines the potential to apply quantum computing to forensics and how it can be used to transform the field through its disruptive capabilities in the evidence collection process, detection of intrusions, modeling of cybercrime, and DNA analysis. It also underrates the dangers that quantum technologies bring to the data security and the urgency of post-quantum encryption technologies. The article presents a blueprint of quantum-driven forensic investigation of the near future by conducting a survey of recent advances and new applications. We combine theory and practice by using datasets such as NSL-KDD, Qiskit simulations, and diagrams of how quantum machine learning models, DNA profiling and intrusion detection systems are used. Pattern matching in the DNA profiling algorithm with quantum computing is determined to have a time complexity of O(n) in the application of the Grover algorithm and O(n) of the corresponding classical algorithm. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Hybrid GNN-Driven Framework for Intelligent Malware Detection and Cryptojacking Prevention in Heterogeneous Cloud Environments
Cloud environments are increasingly targeted by cryptojackers who use the computers processing capabilities for mining cryptocurrency without authorization. This research aims to enhance the security features that protect against cyber attackers by implementing deep learning techniques that help to detect anomalous behaviors in the cloud through analysis of data from typical system transactions. The hybrid HGCN-SIEM Fusion architecture for cryptojacking prevention and malware detection incorporates four types of Graph Neural Network (GNN) approaches: GCN, GAT, GIN, and GraphSAGE. The proposed technique achieves superior malware detection accuracy compared to all baseline models. After experiments on the standard SoK cryptojacking malware dataset, GAT and GraphSAGE demonstrated an accuracy average of 97.5%, GCN and GIN achieved similar accuracy, with an average score of 95.5%. The HGCN-SIEM model outperforms with an optimum accuracy of 98.8%, ensures low latency, and provides a well-balanced mix of rapid attack detection and the best utilization of the network bandwidth. SHA-256 is used to hash all process, instance, and event identifiers to protect privacy and ensure distinct, impenetrable node representations. Graph sampling, edge pruning, and adaptive batching are used to manage computational scalability in heterogeneous cloud networks, which reduces latency, increases throughput, and optimizes resource utilization during inference. This research work points out those GNN architectures that combine different node types that are extremely useful for security monitoring and malware detection in various network settings, demonstrating reliability and practicality in cybersecurity contexts. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Real-Time Network Monitoring: Integrating Machine Learning and Custom Packet Sniffer Using Python
The growth in network traffic and the increasing complexity of cyber threats necessitate robust systems for detecting anomalies that indicate security breaches. This research presents a methodology for finding anomalies in packets sent when the connection is established. It uses a machine learning model and a packet sniffer. It captures Transmission Control Protocol (TCP), User Datagram Protocol (UDP), IPv4, and Internet Control Message Protocol (ICMP) segments to predict if any anomalies are present (Sanders in Practical packet analysis: using wireshark to solve real-world network problems, No Starch Press, San Francisco, 2017). An unsupervised learning model is utilized. The presence of unlabeled data to enhance the real-time prediction using isolation forest model. The data collected by packet sniffer undergoes avoiding null values and encoding addresses, and thus an isolation forest is used so that it predicts if anomalies are present using binary trees. The performance is evaluated on the basis of metrics like accuracy, precision, and F1-score (Goutte and Gaussier in European conference on information retrieval, Springer, New York, 2005). The result illustrates the model is accurate in predicting whether anomalies are present. Future work is focused on enhancing the models capabilities with more protocols and an active defense mechanism. The study addresses real-world deployment challenges especially in heterogeneous environments like IoT-based networks. While isolation forest is getting high accuracy, future research could explore hybrid approaches combining traditional statistical methods with deep learning techniques for enhanced industry applications (Ahmed et al. in J Netw Comput Appl 60:1931, 2016). The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.