Browse Items (16481 total)
Sort by:
-
Deep learning-based diabetic retinopathy detection with advanced image segmentation and transfer learning techniques
Diabetic retinopathy (DR), a dangerous side effect of diabetes, can result in permanent blindness. This work presents a state-of-the-art deep learning-based system that uses retinal images to detect and classify DR early on. Utilizing transfer learning and pre-trained models, the system combines Django, Numpy, and Keras to improve diagnostic precision. It accurately detects DR-affected areas and delivers real-time graphical outputs for prompt medical interpretation and decision-making using the ResNet and Mask RCNN architectures. Simple picture uploads are made possible by the user-friendly interface, which lets Numpy handle data processing and preparation. To improve accuracy and reduce the amount of new data required, the system uses transfer learning and pre-trained datasets. The system's robustness and efficacy are highlighted by its evaluation, which shows its high accuracy with an overall accuracy of 95.55%, precision, recall, and F1-scores above 0.95. The suggested approach provides an affordable, effective, and scalable means of detecting DR early on; it is especially helpful in healthcare settings with limited resources. The technology has the potential to greatly enhance patient outcomes and lessen the toll that diabetic retinopathy has on both individuals and healthcare systems by enabling prompt diagnosis and treatment. 2026 Author(s). -
High-precision lung disease detection and classification from chest radiographs using deep and ensemble neural networks
Chest X-rays are a quick and effective way to diagnose lung diseases. This research developed deep learning models to automatically detect chest X-rays of COVID-19, normal, and viral pneumonia patients. The goal was to evaluate deep learning for automated detection of lung diseases from chest X-rays. The research implemented transfer learning with ResNet101 and EfficientNetB0 architectures using a public chest x-ray database with over 21,000 images across COVID-19, normal, and other pneumonia infection classes. Pretrained ImageNet weights were used to initialize the models before fine-tuning them to classify features in chest X-rays. Data augmentation techniques like rotation, shifting, and flipping were applied to expand the number and diversity of training images. The models achieved exceptional performance with accuracy scores of 93.7% for ResNet101 and 95.3% for EfficientNetB0 on test data. Additionally, an Ensemble model, the combination of the two models, was implemented, achieving an accuracy of 96.4%. The findings demonstrate the capability of Ensemble deep convolutional neural networks for accurate automated classification of chest X-rays for Lung disease. Through data augmentation and transfer learning, high-precision models were developed without needing exceedingly sizeable medical image datasets. These deep learning classifiers could serve as rapid diagnostic decision support systems to identify potential lung disease patients using readily available chest X-rays. Such tools could assist healthcare providers, especially when access to expensive diagnostic tests is limited. 2026 Author(s). -
Advancements and challenges in deep learning for breast cancer screening: A review
Breast cancer continues to be the prevalent cancer on a global scale, playing a major role in the worldwide cancer statistics, the critical role of early detection in reducing death rates is underscored. In the context of breast cancer, screening, deep learning (DL) emerged as a game-changer, providing notable improvements over existing techniques. This review explores the use of DL in analysing images from various sources such as X-rays, ultrasound, magnetic resonance imaging, and biopsies. Additionally, it highlights DL's potential to pre-screen for cancer by integrating diverse data, including demographic information, biological markers, and meta-analytical risk assessments. The analysis reveals that deep learning frameworks, especially those optimized with feature selection techniques, attain the minimal false-negative rates, effectively distinguishing between patients with and without cancer. Notably, DL models demonstrate lower prediction uncertainty compared to traditional machine learning, as shown by reduced standard deviations in performance metrics. Additionally, the paper proposes a cascade network model that achieves 98.61% classification accuracy and a 98.41% F1 score by segmenting tumours with a UNet architecture and classifying them with a ResNet backbone. Despite these advancements, challenges such as limited annotated data and adaptability to new data domains persist. In response to these issues, the proposed Self AdaptNet leverages innovative self-supervised learning and adversarial techniques to improve the resilience as well as adaptability of BC detection models.AI technology, particularly DL-based systems, has the capacity to completely transform breast cancer screening by improving screening accuracy and reducing observer variability. However, clinical adoption requires standardized guidelines, trustworthy AI practices, and collaboration among researchers, clinicians, and regulatory bodies. 2026 Author(s). -
IoT based hybrid patient health monitoring system
According to the data released by the World Health Organization (WHO). Cardiovascular diseases (CVDs) are still the leading thread worldwide. Every year around 17.9 million fatalities worldwide are attributed to CVDs, making up 31% of overall deceases. Heart attacks and strokes are the main causes account for most of these deaths, and a large fraction happen so early in the age group of under 70. Check-ups are essential to monitor the healthcondition of the elderly people, which generates a considerable challenge to the existing medical field. Subsequently,It's becoming more crucial to diagnose diseases quickly and accurately with an affordable cost. The World's population growth need for a smart and affordable healthcare solutions to reduce the medical costs. Thus, the development of an effective health monitoring system that can quickly identify irregularities in health and provide precise diagnoses based on gathered data is imperative. New developments in cloud computing and mobile technologies have led to the development of a number of cloud-based healthcare products and services. These cloud systems allow for the automatic collection and transmission of medical data to providers from any place, allowing for the network-based delivery of patient feedback. This project's goal is to use the ThingSpeak IoT platform to create and implement an Internet of Things-based patient health monitoring system in order to meet these goals. 2026 Author(s). -
An exhaustive examination of automatic speech recognition
Speech is nature gift to the human being to differentiate with other creatures on the earth. Speech research stands out as one of the most daunting fields amidst numerous challenging research domains. This current analysis of existing literature aims to provide insights for future endeavors within the global speech research community. The study delves into various challenges associated with speech corpora, front-end algorithms aimed at efficient speech representation, and back- end engines tasked with the recognition process. Thirteen speech corpora undergo scrutiny concerning factors such as language diversity, duration, developmental progress, and accessibility. Furthermore, this research review illuminates potent methodologies that foster the extraction of rich features and bolster robust speech recognition capabilities. This gives an idea on how various methods are available to recognize speech in an effective way. 2026 Author(s). -
Modelling and simulation of high-pressure hydrogen storage tank with composite reinforcement
The hydro-carbon fuel disadvantages like cost, pollution and non-renewable source made a way to look for the other energy resources. The carbon neutral fuel hydrogen is one of the promising fuels for all types of locomotives. One of the major challenges is safe fuel filling and storage, since the hydrogen is highly volatile fuel. Based on the travel distance, different environmental conditions, the hydrogen fuel tank subjected to the varying pressure and volume, which needs the cost-effective material for the fuel tank. This paper presents a comprehensive modelling and simulation study of a highpressure hydrogen storage tank reinforced with composite materials. The performance analysis of a hydrogen storage tank with composite reinforcement is conducted and compared to a standard aluminium hydrogen tank. 2026 Author(s). -
Enhanced Cloud Security: Certificateless Public Auditing Using LBMHT for Malicious TPA Detection
Leveraging the Lattice-Based Merkle Hash Tree (LBMHT), the paper presents a certificate-less publicly auditing technique that targets hostile third-party auditors (TPAs) in cloud settings. Without the usual burden of certificate-heavy and certificate-management-prone identity-based or encrypted with public keys structures, this method seeks to improve information safety and integrity. To enhance the effectiveness of the Key Generation Centre (KGC), decrease complexity of space, and optimise storage in the cloud, the suggested solution utilises multi-ciphertext searching using lattice-based, certificate-less verification. The concept guarantees collision-free hashes by using a Merkle Tree structure, which makes it effective for information confirmation. Based on simulation findings, LBMHT is superior to current AES and RSA methods in terms of performance, decreasing executions, encryption, and decryption durations while simultaneously decreasing communication expenses, responding duration, and memory utilisation. The suggested approach is more economical with resources and works well in scaled cloud settings because of its increased accuracy, less effect from malevolent attackers, and improved throughput. At the end of the section, we go over the benefits of the framework, which include reduced utilisation of resources and validated indicators of performance. Compared to Rivest-Shamir-Adleman (RSA) and Advanced Encryption Standard (AES), the suggested LBMHT method has a higher overall accuracy of 99.4 percent. The efficiency of LBMHT in securely organising and analysing data is shown by its great accuracy. 2026 American Institute of Physics Inc.. All rights reserved. -
Data Analytics in Diabetes Treatment: Approaches and Applications
These days, managing diabetes has become particularly more beneficial with advancements in data analytics, using machine learning, predictive analytics, and patient-generated health data to optimize the outcome for patients. This paper explores the latest techniques and innovations in this field, including predictive modelling, wearable technology integration, and artificial intelligence for better personalized care. The study covers various analytical frameworks, compares the performance of multiple machine learning models, and discusses future directions in the integration of data analytics with telemedicine for diabetes care. The following words refer to diabetes management, data analytics, predictive simulation, AI, and algorithms for learning, wearable technologies, patient-generated health data, predictive analytics, continuous glucose monitoring, health informatics, personalized care, health data privacy, predictive algorithms, electronic health records, diabetes complications, telemedicine, feature selection, and model evaluation along with having patient-centric systems and chronic disease management. 2026 American Institute of Physics Inc.. All rights reserved. -
Neonatal Hypoglycemia in the Newborn of a Diabetic Mother: Risk Factors in Prediction and Management
Among the metabolic complications, neonatal hypoglycemia remains the most common complication, especially among infants of diabetic mothers. This paper is to briefly review landmark research studies aimed at assessing the prediction of neonatal hypoglycemia with consideration of maternal glucose monitoring, cord blood C-peptide, and HbA1c biomarkers together with advanced AI-based models. This becomes significant in reviewing the pathophysiology of neonatal hypoglycemia, clinical risk factors, and treatment strategies in view of the predictive value of both the mother and the newborn. Maternal diabetes type, gestational age, and neonatal BMI are some of the risk factors to be evaluated. The rapidly growing role of AI in clinical practice would also be discussed. In addition, an individualized management approach might be advocated to improve the outcome of the newborn by employing appropriate clinical tools coupled with AI-based predictive models. 2026 American Institute of Physics Inc.. All rights reserved. -
The analytical study of public sentiments about nCovid-19 using twitter comments
People's sentiments are the mirror of their cognition, and these sentiments play a very significant role in predicting and shaping one's behaviour. At present the entire world is fighting with this pandemic situation due to the nCOVID-19 outbreak and people are experiencing a variety of sentiments. This research aims to explore various sentiments that people are experiencing during this epidemic. To achieve this objective sentiment analysis was conducted on 30,000 random Twitter comments using R software. Data mining of data was done using three hashtags: #Coronavirus, #Covid19 and #Covid19India. After the analysis, it was found that the nature of people's sentiments about nCOVID-19 is majorly positive. This study also elicits other noticeable patterns of netizen's expression through their comments while combating nCOVID-19. The present research provides insight into the type of sentiments which people are undergoing across the world during this pandemic situation and based on the obtained data, risk prediction can be done, and various awareness programmes can be designed to overcome the present issue which is prevailing worldwide. 2026 Author(s). -
Integrating Vertical Greening Systems for Urban Heat Mitigation and Well-being in Bengaluru's High-Rise Buildings: A Literature Review and Pilot Study
Rapid urbanization in Bengaluru has aggravated the Urban Heat Island (UHI) effect and decreased green space in high-rise developments. This phenomenon creates elevated "heat hotspots"that increase cooling energy demand and impact public health, social equity, and economic sustainability. To mitigate these issues, balcony greening and other Vertical Greening Systems (VGS) are considered nature-based solutions. This research paper integrates a comprehensive literature review of VGS performance with a pilot study examining Bengaluru residents' perceptions. The pilot study comprises a cross-sectional survey of 55 participants (95% CI: 13.2%). Existing literature demonstrates VGS effectiveness in reducing surface temperature by 2-4C and ambient temperature by 1-3C, thereby reducing cooling energy requirements by 15-23%. Survey results indicate high acceptance (80.9%, 95% CI: 68.5-89.7%), with 87.5% (95% CI: 76.0-94.1%) recognizing VGS benefits for cooling and psychological stress reduction. However, maintenance burden (54.5%), structural concerns (25.5%), and native flora scarcity (73.2%) were identified as significant barriers. Chi-square analysis revealed statistically significant associations between acceptance levels and perceived benefits (?2 = 18.42, p < 0.001), indicating strong adoption potential when barriers are addressed. This research paper offers critical insights into tropical high-rise vertical greening perceptions, informing climate-resilient urban development policies for Bengaluru and similar megacities. Published under licence by IOP Publishing Ltd. -
Yaj Bhava in teamwork: whole is greater than the sum of its parts
Purpose The purpose of this study is to extend contemporary teamwork theories by offering a value-based, ethically grounded model for managing teams in the current uncertain challenging environment. Design/methodology/approach A qualitative content analysis of objectively shortlisted relevant verses from the Bhagavad Gita to identify its core principles. Findings Incorporating Yaj Bhava into organizational practices can significantly enhance cohesion and success both at the Individual and Team level. Research limitations/implications This study introduces Yaj Bhava as a value-driven approach that complements and extends existing teamwork theories. It shifts focus from external coordination to internal attitudes like selflessness, shared purpose and ethical contribution. This perspective encourages a deeper exploration of how inner dispositions shape team effectiveness. The framework opens avenues for cross-cultural research and invites the development of new tools to study teamwork through a value-based lens. It also challenges researchers to revisit core constructs such as trust, motivation and goal alignment from a more holistic and ethical standpoint. Practical implications Applying the principles of Yaj Bhava encourages a shift in teamwork practices where prioritizing collective well-being over individual recognition leads to more integrated and sustainable outcomes for organizations and society. Originality/value This unique approach applies Indian Knowledge Systems to modern workplace challenges, bridging ancient wisdom with contemporary organizational needs. 2025 Emerald Publishing Limited -
A Comparative Review of Lossless Text Compression Algorithms: From Classic Techniques to Hybrid Models
The lossless text compression is an essential part of data transmission and storage that allows using resources effectively without losing data integrity. This paper combines the most recent research with the important discoveries of benchmark research that focuses on essential algorithms such as Huffman LZW and Shannon-Fano and advanced hybrid algorithms such as LZW and Burrows-Wheeler Transform (BWT -based algorithms. The trade-offs between compression ratio, speed and adaptability can be observed comparing how algorithmic concepts, operational stages and empirical performance evolve in different datasets. This evaluation ends with recommendations on how to choose algorithms as well as future research recommendations. 2025 IEEE. -
Converging Deep Learning and Cloud Computing: A Scalable and Efficient Approach for Modern AI Infrastructure
Deep learning has proven to be a powerful approach to solving challenging problems, ranging from natural language processing, speech recognition, to computer vision. The proposed model has capable of matching the expanding amount of facts as well as complexity of the algorithms is demands of deep researching methodologies. These demands cannot be met with traditional computing environments. This is why cloud computing technologies have developed, offering a scalable and affordable alternative for executing deep learning algorithms. Cloud computing platforms provide the resources necessary to run deep learning workloads including compute, storage, and networking. This means you no longer have to spend lots of money on expensive hardware and makes it easier for teams and researchers to deploy and train deep learning models faster. Cloud computing has ushered in a new era of deep learning developments by mobilizing the power of specialized hardware, notably GPUs, to accelerate the training and performance of deep learning models. 2025 IEEE. -
High-Performance 15-Level Multilevel Inverter for Renewable and Smart Grid Applications
Multilevel inverters have emerged as a promising solution for improving power quality, reducing switching stress, and enhancing conversion efficiency in renewable energy and smart grid applications. Conventional two-level topologies struggle with high Total Harmonic Distortion (THD), electromagnetic interference, and bulky filters, limiting their suitability for high-power systems. To address these challenges, advanced multilevel architectures have been designed to deliver multiple voltage steps, thereby approximating a sinusoidal waveform with greater precision. This paper investigates a high-performance multilevel inverter with an optimized 15-level configuration that achieves superior harmonic reduction, enhanced voltage boosting, and reduced switching device count compared to conventional alternatives. A detailed switching sequence and operational modes are provided to demonstrate the generation of fifteen distinct output voltage levels. Simulation and analytical results validate the performance in terms of THD minimization, voltage stress reduction, and waveform quality. The proposed configuration is suitable for integration with photovoltaic systems, wind generation, and smart grid applications where reliability, efficiency, and grid compliance are essential. Furthermore, the design demonstrates modularity, scalability, and compatibility with wide bandgap semiconductors, reinforcing its role as a practical solution for modern energy systems. 2025 IEEE. -
Modeling Popularity Evolution with Popularity-Augmented Graphs and Dynamic Bayesian PARAFAC
In recent years, social media has evolved as a significant platform for attracting new clients and customers. Every day, a wide range of new offers and products are shared over the social media platforms for buying, selling, promotions, etc., encouraging more and more social engagement. Therefore, it's important to predict high consumer engagement using past interactions. This study proposes a two-stage framework that integrates Popularity Augmented Social Graph construction with Dynamic Bayesian PARAFAC decomposition. The experiments were conducted on the open-source Behance project dataset, which contains interactions from over 85,000 users across 1,326 projects over 60 discrete time intervals. In the first stage, a Popularity Augmented Social Graph (PASG) is constructed using the popularity information. In the second stage, the graph is represented in tensor form and is factorized using Dynamic Bayesian PARAFAC (DBPF), which models latent relationships across users, content, and time. The performance of the model was evaluated using Mean Relative Error, Mean Absolute Error, Root Mean Squared Error, where it consistently outperformed the baseline methods. The results demonstrate the effectiveness of the proposed framework in providing a robust and scalable solution for popularity prediction in social media platforms. 2025 IEEE. -
Comparative Performance Analysis of Clustering Algorithms for Scalable and Reliable Vehicular Ad-Hoc Networks (VANETs)
Vehicular Ad-Hoc Networks (VANETs), widely used in intelligent transport systems, require effective clustering techniques to maintain network stability, reduce network latency and enhance communication efficiency. This research presents an in-depth analysis of three widely used clustering algorithms: K-Means, Spectral, and Leiden. Efficacy is assessed across different vehicle densities and speeds. The study focuses on examining four primary factors: the modularity of cluster formation, silhouette score, throughput, packet delay and cluster head change rate. The results obtained from the tests indicate that K-Means always sends more data & has the quickest packet delivery which generates the best-shaped clusters to elect CH. This is the best choice for networks with a varying number of cars that change speeds. Leiden does well when there are a lot of cars on the road. It stays stable but changes for huge graphs. Spectral clustering always does worse, with longer delays, less data getting through, and cluster heads that change too much. These findings show that selecting the right algorithm matters when building VANETs that can grow and stay reliable. The study concludes that K-Means is the best choice for cluster formation & electing CH where there is a need for quick responses and lots of data flow. Leiden works well in packed networks that need balanced performance. Spectral clustering does not work efficiently when keeping the network running in real-life vehicle situations at higher density & speed. 2025 IEEE. -
Multi Disease Identification in Tomato Plant using CNN and SVM
Tomato is a major trade crop; it is among the most widely consumed crops in daily life. Crop diseases reduce not only the quality of the crops but also their amount of production, thus, detection and identification of the specific diseases is of great importance. Diseases like the Mosaic virus, Bacterial Spot, and Yellow Leaf Curl Virus infect the tomato plant. The advanced detection and classification techniques are mainly employed in the diagnosis of these diseases. This helps in informing the farmers about the types of diseases that attack their crops. In this study, independent CNN and SVM classifiers built to classify the diseases. The CNN model extracts feature such as color and leaf edges from input images- then, it proceeds to classification. For SVM, PCA is applied for feature reduction in order to enhance performance and accuracy before classification. A dataset sourced from plant village has been utilized to train the network CNN and SVM. The proposed neural network model has been applied to categorize 4 types of tomato leaf conditions: one healthy and three diseased types of tomato leaves. The results show that the SVM approach achieves a classification accuracy of 94.33%, whereas the CNN model has slightly higher accuracy of 95.17%. 2025 IEEE. -
A Study on the Ethics of using Artificial Intelligence in Mental Health Treatment and its Legality
The development of artificial intelligence (AI) has transformed mental health care through offering new and feasible solutions to old assumptions. The moral concerns related to the use of AI in the mental health, however, could not be overlooked. A thorough grasp of how AI can be used throughout the patient journey is essential to advancing AI technology in the realm of mental health and overcoming its present restrictions. To reduce it to three columns, or one dataset, five Facebook datasets were gathered from Kaggle. The preprocessing procedure enhances the dataset's quality by using user tweets. Four datasets about depression were taken from the Kaggle website. After the preprocessing is finished, we will refine four pre-trained BERT models using the Hugging Face package. We will be able to create a predictive model for identifying depression with this method. The effectiveness of our refined BERT models for depression identification was assessed using a number of metrics. Our healthcare system could be greatly enhanced by AI, but we can only realise this potential if we begin addressing the moral and legal issues that currently confront us. 2025 IEEE. -
A Comparative Benchmark of Deep Learning and Classical Models for BLE-Based Indoor Localization
Bluetooth Low Energy (BLE)-based indoor positioning has gained attention as a cost-effective solution for environments where GPS signals are unreliable. Despite advances in ML and DL techniques, few standardized benchmarks exist for comparing models under uniform conditions. This study evaluates seven models - K-Nearest Neighbor, Random Forest, Deep Neural Network, 1D CNN, Long Short-Term Memory, Bi-LSTM, and Transformer - on a publicly available dataset collected across multiple building floors. A preprocessing pipeline was applied to address missing values, refine RSSI signals, and generate temporal features. Performance was assessed using both accuracy metrics (MAE, RMSE) and efficiency metrics such as processing time, and model size. Results show that KNN, Random Forest, and DNN consistently outperformed complex sequential and attention-based models, achieving RMSE as low as 1.297 m. These findings suggest that simpler architectures align more effectively with BLE RSSI data than deeper models. This study establishes a benchmark that can support future work in developing efficient, lightweight, and generalizable indoor positioning systems. 2025 IEEE.