Browse Items (16481 total)
Sort by:
-
Characterizing Context-Dependent Biochar Effects: An ANOVA-Based Study on Soil Properties and Microbial Diversity
Contemporary intensive agriculture has improved food security, but is a detriment to soil health, biodiversity, and long-term sustainability. Biochar is an exciting product derived from the pyrolysis of biomass that possesses great potential to be a soil amendment that can improve soil chemical, physical and biological properties and sequester carbon. This paper summarizes recent international studies (2024-2025) and contains experimental analyses showing how biochar had an effect on soil systems. Considering soil pH, hydrophobicity, porosity, and particle size were emphasized. Our findings indicate that biochar improves soil structure, water retention, nutrient retention, and diversity in microbes, all of which increase crop resilience under abiotic stress conditions. However, there is a context-sensitivity to the utilization of biochar - often changing with soil types, feedstock, pyrolysis, and application rates. By using standardized and characterizing methods in soil characteristics and ANOVA based statistical analysis, this study presents the rationale and insights, opportunities and limitations of biochar as a sustainable soil conditioner. Further, the findings suggest to tailor "designer biochars". It seems plausible that these could be optimized for targeted soil and crop systems, and be a vital tool in developing climate-resilient and sustainable. 2026 IEEE. -
Wired Highways: The Soul of a Smarter City
The "smart city"concept emerged in the 1990s, characterized by extensive globalization, rapid technological advancement, and the emergence of the knowledge economy. Urban planning professionals began considering how to introduce digital technology into existing city infrastructure to create cities that were faster, efficient, sustainable, and better managed. Innovative thinking has moved away from traditional methods of city planning and management, and has started to challenge the planning systems where governments understood complexity from a bureaucratic perspective. In planning smart cities, local governments are largely relying on using information and communication technologies (ICT) to enhance essential city functions like transportation, energy, water, and public safety. Urbanization is becoming increasingly complex, and the global urban population is growing at unprecedented rates, creating a need of urban systems to be smarter, safer, and more environmentally sustainable. Cities are faced with many complex problems, but urban transportation is one of the most daunting. Existing infrastructure is becoming overwhelmed by congestion, increasing volume of traffic, and safety issues. An innovative way of tackling these problems is through Intelligent Transportation Systems (ITS). ITS is the use of ICT including sensors, software, communication networks, and data analytics in transportation systems to increase efficiency, safety, reliability, and sustainability. By using real-time monitoring, predictive analytics, and data-driven decision-making, ITS creates traffic management systems that improve city response time, decreases congestion and accidents, and increases sustainable or environmental-friendly transport choices. This research recognizes the importance of ITS to smart city development, emphasizing. 2026 IEEE. -
Breast Cancer Prediction using a Stacked Ensemble of XGBoost and LightGBM with Logistic Regression Meta-Learning
Breast cancer remains one of the major reasons for cancer deaths in women, which is why it is key to develop and improve diagnostic systems for accurate predictions. Currently, the advent of Machine learning has helped in providing powerful algorithms to achieve advancements in cancer detection. However, the main motivation of this research is to focus on building more complex ensemble architectures, as they are known for significantly improving predictive accuracy, robustness, and generalisation, especially in performing complex tasks such as medical diagnosis. In this research, a Hybrid stacking ensemble was built using two gradient boosting techniques, XGBoost and LightGBM, with a Logistic Regression meta-learner to predict breast cancer and compare their performance with standard classifiers. The Breast Cancer Wisconsin (Diagnostic) dataset, which consists of 569 patient records, was utilised for model training and analysis. The data was preprocessed using Z-score normalisation and stratified 5-fold cross-validation. The machine learning algorithms, such as Decision Tree, Logistic Regression, and Random Forest, were compared with the hybrid model, and the metrics used for comparison were accuracy, precision, recall, F1-score, and ROC-AUC. The proposed hybrid model performed well, achieving a high accuracy rate of 97.37% and a recall rate of 93.00% for malignant cases. McNemar's test (p > 0.05) confirms that this accuracy rate is statistically equivalent to the Random Forest classifier. These findings proved that the proposed model can perform optimally in predicting complex data with the same degree of precision as the standard models. Therefore, the hybrid model can be considered a robust and reliable new alternative for breast cancer prediction. 2026 IEEE. -
A Sentence-Level Risk Estimator for Identifying Hallucinations in Generative AI
Hallucination, defined as the generation of factually incorrect or ungrounded content, represents a critical challenge in large language models and summarization systems. Existing evaluation metrics often operate at the document level and fail to pinpoint erroneous sentences with sufficient granularity. This work introduces Sentence-Level Risk Estimation (SRE), a unified framework for detecting hallucinations at fine granularity by integrating three complementary signals: semantic alignment using BERT-based embedding similarity, QA-based factuality verification through question-answer pair generation and validation, and Natural Language Inference (NLI) entailment assessment using pre-trained models such as DeBERTa-MNLI. These signals are aggregated into a unified Sentence Risk Score (SRS) via weighted calibration. Experimental evaluation on CNN/DailyMail and XSum datasets demonstrates that the proposed method achieves precision of 0.85, recall of 0.75, F1-score of 0.80, and correlation with human judgments of 0.85, representing substantial improvements over existing approaches including FactCC, QAGS, and SummaC. The proposed framework enables AI systems to flag risky sentences for review or regeneration, thereby improving trust and safety in generative applications. 2026 IEEE. -
UWB Radar based Respiratory Rate Detection for Driver
Continuous health monitoring and the early detection of physiological abnormalities play an important role in vehicular environments. In particular, respiration rate and heart rate estimations are crucial for preventing accidents caused by sudden health impairments to the driver. Impulse radio ultra-wideband (IR-UWB) radar provides an effective solution for long-duration and non-invasive respiration rate monitoring. UWB systems offer sub-nanosecond time resolution while operating at low transmitted power levels, making them suitable for continuous monitoring of the human body. UWB pulses possess strong penetration capability, allowing signals to pass through obstacles such as clothing and vehicle seat covers. This paper presents an IR-UWB radar-based framework for estimating respiration rate using a seat-integrated monostatic radar configuration, where UWB signals propagate through the thoracic region from the posterior side toward the lung. Respiration-induced variations in lung geometry and dielectric properties under different physiological conditions result in corresponding changes in the reflected pulses, which can be analysed for respiration monitoring. Furthermore, variations in the antenna reflection coefficient (S11) exhibit noticeable differences under different lung conditions, from which respiration waveforms can be derived. The extracted respiration-related signal is subsequently transformed into the frequency domain using the Fast Fourier transform (FFT), which enables the accurate estimation of the respiration rate. In this paper, the UWB signal for radar communication complies the Federal Communications Commission (FCC) spectral mask from 3.1 - 10.6 GHz to ensure human safety. The results presented in this paper confirm that the proposed UWB Gaussian seventh-derivative IR-UWB Radar combined with FFT-based processing enables reliable respiration rate estimation and is well-suited for continuous in-seat vital sign monitoring in driving environments. 2026 IEEE. -
Predicting Crude Oil Futures using Feed Forward Neural Networks and Technical Indicators: A Comparative Study on WTI and Brent
In the domains of economic management and energy analysis, forecasting the price of crude oil is increasing popularity. It is essential to the facilitating rapid and cost-effective development with improved quality. Accurate prediction of the crude oil market is essential for steady and fast economic development because of its enormous influence on the global economy and society. Moreover, precise crude oil price prediction aids the traders in making accurate decision to maximize profits. In this work, a machine learning method for forecasting future global price data for crude oil is provided based on past data. The proposed model consists of three phases: primarily, historical data of selected crude oil data are gathered and normalized using data normalization technique. Secondly, technical indicators are derived from the crude oil data. Finally, a Feed Forward Neural Network (FFNN) is designed and trained using these technical indicators to forecast the price of crude oil in the future. Daily, weekly, and monthly data from Brent crude oil and West Texas Intermediate (WTI) are used to evaluate the generated model's prediction ability. To find the most effective FFNN configuration, the model's efficacy is evaluated by adjusting hidden layer number and hidden neurons. Performance of the model is also analyzed by varying number of training and testing samples. The experimental outcomes demonstrates that the designed model exhibits excellent performance for both WTI and Brent data. Notably, the model proves to be effective in predicting crude oil prices, when technical indicators are used as input variables. 2026 IEEE. -
Climate Change and Rainfall Variability in Goa: A Hybrid LSTM-Autoencoder based Predictive Approach
Climate change has significantly altered precipitation patterns in coastal regions like Goa, India. Rainfall serves is a critical resource for crop cultivation in many developing countries. Accurate forecasting of rainfall is essential for sustainable planning, agriculture, and disaster mitigation. However, forecasting rainfall is still challenging due to the dynamic and non-linear nature of weather data. The intricate temporal correlations included into the data may be difficult for traditional time series models and machine learning techniques to adequately reach. This demands the use of advanced data-driven techniques capable of identifying these intricate patterns. This paper presents a data-driven approach using a Long Short-Term Memory Auto Encoder (LSTM-AE) to predict rainfall anomalies over Goa. Seven weather parameters are collected, preprocessed, and analyzed to train the LSTM-AE model. Efficacy of the model is assessed by computing MSE, MAE, and R2. Experimental results demonstrates that the proposed model exhibits strong predictive capability. This research contributes to enhancing early warning systems and developing adaptive climate strategies for the region. 2026 IEEE. -
A Deep Learning-Enhanced Adaptive Intrusion Detection Framework for Real-Time Cyber Security Threat Analysis
The rapidly evolving nature of cyber-attacks significantly reduces the effectiveness of conventional intrusion detection systems (IDS) that rely on static rules and signatures. This work presents an adaptive deep learning- based intrusion detection framework designed to maintain reliable performance in real-time environments affected by concept drift. The proposed approach integrates one-dimensional convolutional neural networks (1D-CNN) for local feature interaction learning with a bidirectional long short-term memory (BiLSTM) network to model sequential network traffic behavior. To address evolving attack patterns, a sliding-window-based incremental learning mechanism is employed, enabling continuous model adaptation to recent traffic characteristics. The model is trained using cross-entropy loss optimized with the Adam optimizer, while dropout regularization is applied to reduce overfitting and ensure fast convergence. To enhance transparency and analyst trust, explainable artificial intelligence techniques are incorporated, including SHAP-based feature attribution and an attention mechanism for interpreting temporal dependencies. Experimental evaluation on labeled network traffic data demonstrates stable convergence, consistent detection accuracy under changing traffic conditions, and improved robustness compared to non-adaptive baseline models. These results confirm the effectiveness and practical applicability of the proposed framework for real-time and interpretable cybersecurity intrusion detection. 2026 IEEE. -
Clinical Pattern Mining for Early Detection of Chronic Kidney Disease: A Data-Driven Diagnostic Framework
Early diagnosis of the Chronic Kidney Disease (CKD) is essential to avoid irreversible damage of the kidneys, but it is clear that the traditional threshold-based techniques of the diagnosis are not always able to detect a subtle pattern of biochemical changes, which indicate the early appearance of the disease. This paper provides an interpretable and data-intensive diagnostic model which incorporates clinical state transformation, frequent and contrast pattern mining, and phenotype-based clustering to reveal hidden signs of CKD progression. Continuous laboratory variables are discretized into clinically meaningful states, enabling transparent rule extraction and comparative analysis between CKD and non-CKD cohorts. The mined contrast patterns reveal distinctive early-stage abnormalities, including mild creatinine elevation, reduced urine specific gravity, albuminuria, and increased urea levels, which consistently differentiate diseased patients from healthy controls. Furthermore, K-means clustering identifies three clinically relevant renal phenotypes corresponding to early, moderate, and advanced biochemical deterioration. Sensitivity and comparative analyses demonstrate the robustness of the extracted patterns across varying support thresholds and against standard machine learning classifiers. The proposed framework offers a clinically interpretable and computationally efficient decision-support tool for early CKD detection and patient stratification using routinely collected clinical data. 2026 IEEE. -
Yoga Hand Mudra Classification using Zernike Moments
This study introduces a novel approach to classifying yoga hand mudras using Zernike Moments, emphasizing their relevance in balancing the Tridoshas - Vata, Pitta, and Kapha, as per Ayurvedic philosophy. A dataset of 1,200 images was collected from yoga practitioners in Bangalore, representing six mudras in both correct ("RIGHT") and incorrect ("WRONG") positions, ensuring a balanced distribution. Zernike Moments were used to extract rotation-invariant shape features from the images. However, due to their lack of scale invariance, Scale-Adaptive Zernike Moments (SAZM) were introduced by incorporating a scaling factor to normalize object size. Image preprocessing involved resizing, Gaussian blur for noise reduction, and normalization. Feature extraction was followed by labeling, scaling, and classification using a Support Vector Machine (SVM). Comparative analysis showed that standard Zernike Moments achieved an accuracy of 55.32%, serving as the baseline. In contrast, SAZM significantly improved classification accuracy to 71.01%, highlighting the importance of scale invariance in gesture recognition. This work demonstrates how computational techniques can complement traditional knowledge, offering a promising direction for yoga-based wellness solutions. While SAZM showed superior performance, future work will address challenges like complex transformations to enhance model accuracy and applicability. 2025 IEEE. -
Event-Triggered Polynomial Model Predictive Control for Multi-Agent Navigation
This paper proposes an event-triggered polynomial model predictive control method for collision-free point-to-point multi-agent navigation. In this control method, each control input to each agent is a polynomial whose coefficients are updated in an event-triggered manner. For each agent, we design an event-triggering rule that guarantees non-Zeno behavior of inter-event times. At each event, the controller updates the coefficients of the polynomial control law corresponding to a subset of agents by solving one or more finite horizon optimization problems. We also ensure feasibility of the optimization problems solved at each event. Through numerical simulations, we illustrate the results and compare the proposed method with other existing methods. 2025 IEEE. -
Machine Learning and Word Representation Techniques in Medical Transcription Data Analysis
Dermatology is the branch of medicine that deals with the diagnosis, treatment, and prevention of skin diseases. Dermatological diseases can be difficult to diagnose, treat, and manage because there are several skin conditions, each with its unique set of symptoms and causes. Underlying medical conditions, environmental causes, or hereditary characteristics can cause complex skin problems. Furthermore, because skin problems can present in a variety of ways, obtaining an appropriate diagnosis and efficient treatment may be difficult. Treating dermatological disorders is a difficult endeavor. This article proposes an integrated model to assist people in understanding and discussing the nature of dermatology. This model's capabilities include text pre-processing, audio-to-text translation, named entity recognition (NER) for extracting crucial information, and text clustering and classification based on content. The necessity for precise and efficient analysis of large amounts of text data, notably the identification and standardisation of abbreviations and the extraction of relevant information, has been identified as a problem in dermatology and medical transcription. By grouping similar cases, clustering can make it easier to spot patterns and trends in dermatological disorders. However, classification can help automatically group text data into pre-established categories, such as various kinds of skin conditions or treatments. These methods simplify data analysis, increase accuracy, and assist healthcare professionals in reaching accurate conclusions regarding patient care. This article explores the partitioning algorithm for clustering, while logistic regression is used in classification. The model analysed in this article helps dermatologists and patients understand and manage skin problems. 2025 IEEE. -
Advancements in Hand Gesture Technology for Enhancing Accessibility in Disability Assistance
Hand gesture recognition technology has become a crucial innovation in assistive technology, providing enhanced accessibility and independence for individuals with disabilities. In order to enhance communication, mobility, and rehabilitation support, this paper investigates the development and integration of AI-powered gesture recognition systems with wearable devices, augmented reality (AR), and the Internet of Things (IoT). The adaptability and ease of use of the current solutions, such as sign language interpretation and smart prosthetics, are severely limited. We propose a novel framework that combines cloud-based data storage, haptic feedback mechanisms, and real-time AI processing to create a highly responsive and personalized user experience to fill in these gaps. The research focuses on accuracy, responsiveness, and ease of use during its comprehensive analysis of prototype testing data and user feedback. The system is able to continually improve the accuracy of gesture recognition and adapt to the requirements of each user by making use of deep learning algorithms. The study also emphasizes the possibility of incorporating brain-computer interfaces (BCIs) for improved control and responsiveness. By providing individualized therapeutic exercises and real-time feedback, our findings suggest that incorporating gesture-controlled interfaces into rehabilitation programs can significantly benefit stroke patients and individuals with motor impairments. Gesture-based smart home control is also made possible by IoT connectivity, which makes life easier for people with limited mobility. An assessment of the system's impact, obstacles to widespread adoption, and potential future directions for improving AI models and making them more affordable are presented at the study's conclusion. The goal of this study is to help close the digital divide for people with disabilities and contribute to the ongoing development of accessible technology. 2025 IEEE. -
From Prediction to Action: Counterfactual Explanations and Ensemble Learning for Explainable Maternal Health Risk Modelling
Maternal health is critical to women's well-being, particularly during pregnancy, delivery, and postpartum. Early prediction and prevention of health risks are essential for reducing complications and improving outcomes. This research introduces a stacking ensemble model for maternal health risk prediction, combining the strengths of Random Forest, XGBoost, and Gradient Boosting with XGBoost as the meta-model. The ensemble approach enhances accuracy and reliability, achieving a classification accuracy of 91.13%, with precision, recall, and f1-scores exceeding 85% across all risk categories.Beyond accurate prediction, this study emphasizes model interpretability through Diverse Counterfactual Explanations (DiCE), an Explainable AI (XAI) method that provides actionable insights for risk reduction. Counterfactual analysis identifies the minimal changes needed in the patient features to shift a high- or medium-risk classification to low-risk, offering clinically relevant recommendations. These counterfactuals are generated to ensure feasibility, preserving physiological plausibility and practical applicability for healthcare professionals. This work bridges the gap between black-box machine learning models and actionable decision-making by integrating predictive power with explainability, supporting more transparent and patient-centric maternal health interventions. 2025 IEEE. -
Predicting Liver Injury Risk from Chemical Properties and Drug Label Information Using Machine Learning Models
This research aims to create a drug-induced liver injury (DILI) severity prediction system based on machine learning to aid healthcare professionals in safety assessment. FDA's Liver Toxicity Knowledge Base supplied a drug dataset of 1042 drugs, and later, after pre-processing and API data extraction, each drug was defined by 16 chemical features such as molecular descriptors and pharmacokinetic properties. To improve uniformity and get quality input for training, data preparation involved correcting missing values, encoding categorical values, and normalising numerical data. Various machine learning models were trained and evaluated to forecast the levels of DILI severity, i.e., Random Forest, Gradient Boosting, and XGBoost. The importance of features was approximated for identifying the predictors that impacted the most. The best overall performance was recorded for XGBoost, and it had 81% accuracy when it was evaluated. Its acceptable discrimination was established for mild, moderate, and severe cases. The aptness of being applied to the medical sector is demonstrated by drastically lowering the principal misclassifications, especially from mild to severe. The application of machine learning in improving medicine safety assessment and reducing risks associated with pharmaceutical development is illustrated here. 2025 IEEE. -
Smart Autonomous Robot for Efficient Hospitality Service
The hospitality industry is constantly striving to deliver excellent guest experience in the form of timely service and quality food. However, increasing customer expectations have been challenging the industry in the form of workload management, the recruitment of skilled personnel, and the management of operating expenses. Faced with these challenges, companies are embracing state-of-the-art technologies, and mobile robots have been viewed as a potential solution. This paper presents the concept design of a state-of-the-art autonomous robot for food delivery in hospitality establishments. Inspired by robots such as Amazon's Kiva Robots, the robot uses camera modules, path finding algorithms, and sensors to navigate through dynamic spaces while avoiding furniture and moving guests. Unlike warehouse robots, restaurant robots need to learn to adapt to uncertain environments while maintaining the friendly ambiance. By automating routine tasks such as food delivery, the robot allows staff to focus on delivering personalized customer service. Its technical features consist of environmental monitoring camera modules, path-finding algorithm sensors for obstacles detection. It adapts to dynamic environmental conditions for efficiency and safety. Innovation increases operational efficiency, saves labor costs, and improves food quality. 2025 IEEE. -
Approaches To Improve Performance of K-Means Clustering
In this research, we present an enhanced K-Means clustering approach utilizing Neural Engine processors integrated within distributed smartphone networks. Each smartphone runs the K-Means algorithm locally using its Neural Engine to compute centroids efficiently, and these local centroids are then combined to form global clusters on a cloud server. Our implementation significantly reduces computation time while maintaining high clustering accuracy. Experimental evaluation on large datasets demonstrates improved performance over traditional K-Means, proving its suitability for big data analytics in healthcare, IoT, and smart mobile applications. This approach ensures faster processing, lower energy consumption, and effective resource utilization within distributed environments. Further, the proposed method addresses challenges in data privacy by performing local computation and only sharing centroid information. The results indicate potential for scalable clustering solutions in real-time scenarios, opening new directions for edge-cloud integrated machine learning frameworks that harness device-level AI accelerators for complex data-driven tasks efficiently. 2025 IEEE. -
Fortifying Networks based AI Models for Early Vulnerability Detection
Early identification of network vulnerabilities is now essential for protecting sensitive data and guaranteeing system resilience due to the increasing complexity of digital infrastructures. An extensive analysis of artificial intelligence (AI) models intended for the early identification of network vulnerabilities is presented in this research. The study examines current approaches, assesses their efficacy, and pinpoints research gaps while drawing on ideas from recent studies and cutting-edge academic research tools. The results show how AI has the ability to revolutionize cybersecurity tactics and point to new avenues for improving vulnerability detection systems. 2025 IEEE. -
Early Detection and Analysis of Potato Leaf Diseases Using Deep Learning based CNN Models
Potato diseases pose a significant threat to global agricultural productivity, leading to severe economic losses. Early and accurate disease detection is crucial for effective disease management and improved crop yield. This research explores deep learning techniques for automated potato disease prediction using convolutional neural networks (CNNs). A large dataset of potato leaf images is used to train and validate the model, ensuring robustness and accuracy. The proposed deep learning model efficiently classifies common potato diseases, such as late blight and early blight, with high precision. Performance evaluation metrics, include accuracy, The integration of deep learning in disease prediction minimizes the reliance on manual inspection, providing farmers with a cost-effective and scalable solution. Additionally, we analyze the impact of transfer learning and data augmentation on model performance. The results highlight the potential of AI-driven approaches in precision agriculture, offering real-time disease diagnosis and early intervention strategies. This research contributes to the advancement of smart farming technologies, ensuring sustainable crop protection and food security. Future work will focus on optimizing the model for real-world deployment through mobile applications and IoT-based systems. 2025 IEEE. -
Ransomware Detection using Dynamic Behavior Monitoring based on Entropy Analysis and Frequency Analysis
Cybersecurity faces mounting challenges due to the proliferation of ransomware, a sophisticated form of malware that encrypts user data, rendering it inaccessible unless a ransom is paid. Traditional detection systems often fail to counteract evolving threats effectively, creating an urgent need for innovative approaches. Introducing a novel hybrid framework for ransomware detection within IoT ecosystems, integrating entropy and frequency analysis with machine learning models, including Decision Trees (DT) and Random Forests (RF). Data augmentation techniques were employed to generate synthetic data, bolstering the models' ability to generalize across diverse scenarios. Experimental results demonstrated superior performance of the DT classifier, achieving an accuracy of 98.89% and an F1-score of 98.81%. The proposed framework is optimized for real-time ransomware detection, leveraging dynamic analysis to monitor live system behaviors. This integration ensures a proactive defense mechanism against emerging ransomware variants. Future research directions include expanding real-time capabilities, enhancing cross-layer detection, and for collaborative threat intelligence. This work represents a significant advancement in ransomware detection methodologies, offering robust, adaptive, and scalable solutions to mitigate one of cybersecurity's most pressing threats. 2025 IEEE.