Depression remains one of the most prevalent mental health conditions worldwide, yet most automated detection systems focus on English text and overlook the multilingual, multimodal nature of social media communication. This limitation is particularly significant in India, where users frequently write using Hinglish and other Hindi variants, posing substantial challenges for standard language processing pipelines. This study introduces a time-aware multilingual multimodal framework for detecting depression-related expressions in social media posts collected from X (formerly Twitter). The model jointly learns from text, images, and posting time, enabling it to capture how users express themselves across languages and modalities. To support this work, we construct and annotate a dataset of 15,739 posts from 3,471 users covering four linguistic varieties: English, Roman Hindi, Devanagari Hindi, and Hinglish. The contributions of this work are twofold: it presents a multilingual multimodal dataset for Indian social media depression-related content detection, and it provides a systematic evaluation of encoding and fusion strategies, achieving an F1-score of 0.782 with cross-lingual transfer capability demonstrated on a Bengali corpus. All outputs are intended as a content-screening aid and do not constitute clinical diagnosis.

The ability to recognize users’ activities within their homes is essential for enabling assisted living and proactive health monitoring through sensor-based systems. However, in real-world smart home deployments, activity distributions are highly imbalanced: routine daily activities dominate the data, while critical yet infrequent activities remain under-represented. This imbalance significantly limits the generalization capability of conventional machine learning models. In this paper, we propose a deep learning architecture specifically designed to address class imbalance in streaming sensor data. The proposed framework employs a dual-path feature extraction mechanism that integrates hand-crafted statistical features (HCF) with high-level features (HLF) learned using Convolutional Neural Networks (CNNs). The resulting multimodal feature representations are then processed by an ensemble of temporal recurrent architectures, including (i) unidirectional Long Short-Term Memory (LSTM), (ii) Bidirectional LSTM (BiLSTM), and (iii) cascade LSTM layers. To further mitigate the effects of class imbalance, we introduce a specialized focal loss function, denoted as LDLA, which dynamically down-weights the contribution of majority classes during training. Extensive experimental evaluations conducted on five benchmark datasets from the Center for Advanced Sensors and Autonomous Systems (CASAS) demonstrate that the BiLSTM model combined with the proposed focal loss consistently outperforms traditional classifiers. Specifically, it achieves an improvement of 17% to 32% in overall accuracy while maintaining high precision and recall for minority activity classes.

Internet of Things (IoT) enabled Wireless Sensor Networks (WSNs) is not only constitute an encouraging research domain but also represent a promising industrial trend that permits the development of various IoT-based applications. These applications span a wide range from industry to education, and from military to agriculture. The IoT device plays a significant role in various IoT-based networks, and the functioning of such network depends upon the battery power. Once the devices are deployed in the hostile environments, replacing batteries becomes impractical. Despite a plethora of research addressing this challenge, IoT networks still face issues. In this paper, a genetic algorithm based data monitoring and controlling method using IoT enabled WSNs is proposed by using movable sinks in IoT enabled HWSNs (OptiGeA). The OptiGeA protocol is designed for the election of cluster heads (CHs) by incorporating factors such as density, distance, energy and heterogeneous node capacity into its fitness function. The investigation of OptiGeA is conducted with single sink, multiple static sinks and multiple movable sinks provide an unbiased comparative assessment. The novel deployment technique and multiple mobile sinks approaches are proposed to reduce the transmission distance between the sink and CH during system operation and address hotspot issue. It is evident that the OptiGeA protocol shows an increment of 10.44

This study delves into advanced feature selection methodologies for enhancing Android malware classification. GHR-Optimizer is introduced as an innovative feature selection approach combining Grey Wolf Optimization, Hill Climbing, and Random Forest Classifier method. The approach selects features from a hybrid dataset and is evaluated across machine learning, deep learning, and ensemble frameworks. A detailed comparative analysis is conducted, contrasting GHR-Optimizer with static and dynamic feature sets as well as traditional filter and wrapper-based methods. The implementation of the GHR method demonstrated superior performance, particularly when evaluated with diverse datasets such as KronoDroid, which achieved exceptional accuracy and balance in classification metrics. When integrated with the Random Forest classifier, the GHR-Optimizer achieves an accuracy of 98.40%. These findings underscore GHR-Optimizer's superior performance in boosting classification accuracy and robustness, highlighting its pivotal role in advancing feature selection strategies within the domain.

Network security remains a critical concern in today’s fast-paced tech world due to the dynamic challenges posed by high-speed networks. Despite using various security tools like firewalls and intrusion detection systems, managing multiple threats simultaneously is time-consuming. Existing vulnerability assessment tools struggle with complex attack scenarios, particularly those involving chained attacks relying on CVSS metrics. Thus, there’s a pressing need for a new strategy to strengthen network security against evolving threats. Attack graphs emerge as a powerful solution, especially for chained attacks, offering better ranking of attack paths and insights for system administrators. This study proposes a new approach, shifting focus from individual vulnerability severity to the overall severity of hosts or systems. The methodology determines system severity considering inherent vulnerabilities, along with exploit time and cost function. This shift allows a more comprehensive understanding of system security, addressing the complexities of chained attacks. The proposed methodology centres on three key elements: calculating average exploit time, modifying the CVSS v3.1 framework based on exploit time, and calculating a cost function for prioritizing attack paths. Results highlight the need to address 33% of existing vulnerabilities, focusing on 3 out of 9 hosts to enhance overall network security.

Around 280 million people across the world live with depression, making it one of the most common mental health concerns today bib42. Early detection is one of the most effective ways to support those in need and prevent their condition from worsening. Social media lets us see people's daily activities and feelings in today's digital world. More people express their genuine opinions online than in clinical settings. These websites are therefore helpful in learning about mental health trends. However, most previous studies have examined only English text and ignored the variety of languages and media people use on social platforms. This gap is evident in India, where users often write in Hinglish (a natural mix of Hindi and English), which brings new linguistic challenges. To bridge this gap, our study introduces a time-aware multilingual and multimodal framework for detecting signs of depression from social media posts collected from X (formerly Twitter). The model develops a deeper insight into how users behave and express themselves by learning from text, images, and posting times. The results of our experiments indicate that the model performs well, consistently across runs, with an F1-score of 0.79 and an AUC of 0.74, outperforming all text-only or single-language baselines. These results suggest that combining behavioural, visual, and textual cues improves the accuracy and cross-linguistic flexibility of depression detection. This is the first study to examine multilingual and multimodal depression detection using actual Indian social media data. This study shows the value of multicultural research and offers a valuable framework for developing tools that can facilitate online mental health monitoring.

Recent advances in quantum technologies are rapidly pushing the boundaries of computational capabilities, with the promise of significant speedups across a variety of applications. However, building large-scale monolithic quantum processors remains an immense engineering challenge, motivating the pursuit of alternative paradigms to realize practical quantum advantage. One such paradigm is Distributed Quantum Computing (DQC), which interconnects multiple quantum nodes via quantum networks to collectively perform complex computations. While Circuit-Based Quantum Computing (CBQC) has traditionally been the dominant model, Measurement-Based Quantum Computing (MBQC), with its one-way computation framework and pre-shared entangled cluster states, introduces distinctive benefits for distributed architectures. Importantly, within DQC, MBQC offers enhanced security over CBQC by naturally supporting protocols such as Blind Quantum Computation (BQC) and secure delegated processing. This study offers an application-oriented perspective on the role of MBQC in DQC, highlighting concrete industrial use cases alongside key implementation challenges. By bridging theoretical potential with practical deployment considerations, it aims to accelerate progress toward secure and scalable distributed quantum systems.

Over the past decade, we can see a notable increase in the use of multimodal posts (text + images) on social media platforms. In such multimodal posts, images themselves possess the property of multimodality, which means they exhibit different kinds of content (textual/non-textual/both). In the past, limited research has been devoted to extracting the sentiment of users from multimodal images. The main objective of this paper is to categorize multimodal social media political images into distinct classes based on the content they are displaying and subsequently extract the user’s political views from the images. This research proposes a hybrid approach named textual and non-textual image sentiment analysis (TaNTISA). TaNTISA (TaNT + ISA) is a combination of two tasks: (1) textual and non-textual image classification (TaNT) and (2) image sentiment analysis (ISA). TaNTISA contains three modules that perform the sentiment analysis of multimodal images. The first module is the transfer-learning-based textual and non-textual image classifier named TaNT. TaNT is the main module of this research work, which classifies an image into textual, non-textual, or combined image classes with an accuracy of 94

Pancreatic cancer is a highly malignant disease with a low survival rate. The primary reason for its high mortality is the absence of early-stage symptoms and the lack of effective early diagnostic systems. The late-stage diagnosis further exacerbates its mortality rate. Supervised artificial intelligence (AI) techniques have recently demonstrated promising results in earlystage diagnosis. However, a significant challenge with supervised AI methods is their reliance on large, labeled data sets, which are challenging to obtain in the medical field. In this work, the authors investigate the potential of semi-supervised AI techniques to address this limitation. The performance of supervised deep learning classification models is evaluated quantitatively using smaller training datasets. In addition, the authors explore the integration of contrast loss functions with existing AI models to enhance the diagnosis of pancreatic cancer using limited data. The performance of the models is analyzed using three publicly available datasets: BTCV, NIH, and MSD. Experimental results indicate that incorporating contrastive loss with classification loss can significantly improve the diagnostic accuracy of deep learning models, even with a reduced number of training data points.

Detecting plant diseases early and accurately is really important to prevent crops from failing and to make sure we have enough food. This paper looks at using a mix of computer programs that learn, along with "Explainable AI" (XAI) – which helps us understand how the programs make decisions – to get better at finding diseases and to make it clearer how the detection works. Powerful computer programs that are good at analyzing images (called CNNs) were trained on a set of pictures of leaves. Then, the predictions from those programs were combined using a method called "soft voting." This combined approach was very accurate (96.06%), and it did a better job than any of the individual programs. To deal with the fact that it's often hard to understand why these programs make the choices they do, Grad-CAM is used. Grad-CAM creates visuals that show which parts of the images the programs focused on when they were making their decisions. The results show that this combined method is effective at classifying plant diseases, and Grad-CAM is helpful in explaining how the programs reached their conclusions.

The escalation of Internet of Things (IoT) devices has led to increased data generation at the network edge that has burdened the cloud infrastructure in terms of handling and processing of data. This has led to the rapid adoption of fog computing because of its ability to bring computation and storage closer to the edge and support for real-time applications and services by reducing latency. One of the foremost challenges in the fog computing arena is minimizing turnaround time. This research paper proposes a Modified Levy Flight Firefly Algorithm (MLFFA) to optimize task scheduling for fog computing environments. Specifically, the objective is to minimize the turnaround time of tasks. Moreover, genetic operators like crossover and mutation are also employed to achieve an optimal balance between exploration and exploitation. Experimental observations undertaken show that the proposed method improves the average turnaround time by 55%, 22%, and 13%, average waiting time by 59%, 45%, and 37%, average energy consumption by 19%, 7%, and 4%, and average failure rate by 50%, 28%, and 7% compared to the existing studies, namely Load Balancing and Optimization Strategy (LBOS), Technique for Resource Allocation and Management (TRAM), and Fuzzy Golden Eagle Load Balancing (FGELB), respectively.

With the rapid advancement of Internet of Things technology, the field of fog computing has garnered significant attention and hence become a workable processing platform for upcoming applications. However, compared with vast computing capability of the cloud, the fog nodes have resource constraints, are heterogeneous in nature, and highly distributed. Due to the growing demand as well as diversity of applications, the nodes in a fog network become overloaded, which makes load balancing a prime concern. In this work, a load balancing aware task selection and migration approach is proposed comprising two algorithms to select and place tasks from multiple overloaded nodes to suitable destination nodes. The Selection algorithm determines the tasks that should be migrated from overloaded nodes. Placement algorithm focuses on finding a near optimal solution by applying modified binary particle swarm optimization. Specifically, the objective is to minimize execution time and transfer time of tasks. Simulation studies conducted on iFogSim prove that the suggested approach outperforms the existing approaches in terms of task execution time, task transfer time, and makespan.

Background and Objective: Artificial intelligence (AI) methods coupled with biomedical analysis has a critical role during pandemics as it helps to release the overwhelming pressure from healthcare systems and physicians. As the ongoing COVID-19 crisis worsens in countries having dense populations and inadequate testing kits like Brazil and India, radiological imaging can act as an important diagnostic tool to accurately classify covid-19 patients and prescribe the necessary treatment in due time. With this motivation, we present our study based on deep learning architecture for detecting covid-19 infected lungs using chest X-rays. Dataset: We collected a total of 2470 images for three different class labels, namely, healthy lungs, ordinary pneumonia, and covid-19 infected pneumonia, out of which 470 X-ray images belong to the covid-19 category. Methods: We first pre-process all the images using histogram equalization techniques and segment them using U-net architecture. VGG-16 network is then used for feature extraction from the pre-processed images which is further sampled by SMOTE oversampling technique to achieve a balanced dataset. Finally, the class-balanced features are classified using a support vector machine (SVM) classifier with 10-fold cross-validation and the accuracy is evaluated. Result and Conclusion: Our novel approach combining well-known pre-processing techniques, feature extraction methods, and dataset balancing method, lead us to an outstanding rate of recognition of 98% for COVID-19 images over a dataset of 2470 X-ray images. Our model is therefore fit to be utilized in healthcare facilities for screening purposes.

Finding the necessary web service is becoming a more difficult task as a result of the web services' rapid growth in repositories. It has increased the need for effective algorithms for classifying web services. The web service recognition process is enhanced when related web services are clustered or grouped together in service repositories, as this saves search time and space. In order to use vector space to depict web services, numerous distinguished In this field of study, researchers have utilized the Term Frequency - Inverse Document Frequency (TF-IDF), Length Feature Weight (LFW), or Deep learning approach. The TF-IDF or LFW approach generally has a number of drawbacks, such as (1) not being able to determine the semantic meanings among synonyms, antonyms, etc., and (2) Producing a large number of sparse vectors for words that are in the document but not in the corpus used to train the W2V model or unknown words. These factors also contribute to the classification algorithm's declining performance. In this work, we suggest various methods, predicated on Word2Ve for vectorized representation of service followed by classification algorithms: W2V + Naive Bayes, W2V + Random Forest, W2V + SVM, W2V + KNN, and W2V + Logistic Regression. The suggested method, as opposed to TF-IDF or LFW, which only takes into account one weight, helps extract additional information about a word from a web service and stores it in a vector. The suggested methods are used on two datasets of actual web services, and the accuracy, precision, recall, F1-score, and other standard measurement criteria are used to gauge the performance. The suggested approach's outcomes are contrasted with the K-Means(K) clustering LFW representation method. The suggested method performs better than the LFW+ K method, according to the results, and W2V+SVM produces the best outcome.

Recent advancements in the field of the Internet of Medical Things (IoMT) have enabled the real-time monitoring and treatment of patients with communicable infectious diseases while minimizing human intervention. However, IoMT devices face challenges, such as unbalanced energy consumption, memory constraints, computation power, and low latency, which can deter the efficient transfer of patient monitoring data. Thus, there is an urgent need to establish an energy-efficient infrastructure for IoMT devices to remotely monitor and collect data on communicable diseases. For this, a genetic algorithm (GA)-based dynamic transmission of data for communicable diseases in the IoMT environment is proposed in this article. The energy utilization of the IoMT is enhanced by considering the GA evolutionary processing based on the dynamic sensor range. The proposed work incorporates a periphery of the fixed area for deploying the IoMT devices to settle the energy hole problem. Multiple sinks and direct information collection concepts are also introduced which further improve the performance and reduce the movement of data packets. The proposed protocols not only optimize energy usage but also provide a robust approach for massive data collection and communication.

The Indian General Election of 2019 is proof of the rise in political content on social media platforms, reflecting user attitudes and behaviors towards the electoral process. This paper examines the collective behavior of Twitter users during the election period and assesses the feasibility of election prediction through social media analysis in India. This work presents weekly sentiment analysis on election-related Twitter content. It analyzes how social media content can be used as a reliable factor of public sentiment during elections and has recourse to be a predictive parameter alongside other factors. The simulation results show that VADER outperforms various state-of-the-art techniques with a mean absolute error of 1.345 for candidate-wise predictions, significantly outperforming Textblob (5.445), Pattern analyzer (5.135), and pre-trained BERT (28.335) while yielding similar results for party-wise and alliance-wise analyses.

In modern agriculture, deep learning has become increasingly essential to identify plant diseases in the field of plant disease identification with leaf images, where convolutional neural networks (CNN) are garnering a lot of interest. Plant disease iden-tification and prevention are essential for producing healthy crops and maintaining farmers' livelihoods. This study uses leaf images to examine the effectiveness of deep learning (DL) algorithms for multi class plant leaf disease classification. The training data was expanded using data augmentation techniques, which eliminated the need for further image collecting. This paper uses three models namely InceptionV3, VGG16, and EfficientNetBO that have been compared using the plant Village data set. The three models that achieved the highest accuracy were EfficientNetBO (98.51 %), InceptionV3 (97.78%), and VGG16 (95.37%) for plant detection, and EfficientNetBO (97.58%), InceptionV3 (96.83%), and VGG 16 (94.35 %) for plant detection. Notably, EfficientNetBO performed better for disease classification than the other two algorithms. Farmers are able to take prompt action to stop the disease's spread and reduce crop losses since the technology can deliver fast and precise results.

With accelerated advancement of web 2.0, developers generally describe the functionality of services in short natural text. Keyword-based searching techniques are not an efficient way of discovering services from repositories. It suffers from vocabulary problems. Latent Dirichlet allocation (LDA) with word embedding techniques is widely adopted for efficiently extracting latent features from the service descriptions. However, LDA is not efficient on short text due to limited content and inadequate occurring words. The word vectors generated by word embedding techniques are of finer quality than topic modeling techniques. Gibbs sampling algorithm for Dirichlet multinomial mixture (GSDMM) model gives better results on web service description documents because it provides one topic corresponding to one document. In this paper, we evaluate the performance of GSDMM model with word embeddings and propose WV+GSDMMK model. The proposed model improves service-to-topic mapping by determining semantic similarity among features. K-means clustering is applied on service to topic representation. Results are evaluated on five real-time datasets based on intrinsic and extrinsic evaluation measures. Experimental results demonstrate that the proposed method outperforms other baseline techniques, and the accuracy score is also increased by 5%, 18%, 3%, 4%, and 6% on datasets DS1, DS2, DS3, DS4, and DS5, respectively.

Abdominal organs play a significant role in regulating various functional systems. Any impairment in its functioning can lead to cancerous diseases. Diagnosing these diseases mainly relies on radiologists’ subjective assessment, which varies according to professional abilities and clinical experience. Computer-Aided Diagnosis (CAD) system is designed to assist clinicians in identifying various pathological changes. Hence, automatic pancreas segmentation is a vital input to the CAD system in the diagnosis of cancer at its early stages. Automatic segmentation is achieved through traditional methods like atlas-based and statistical models, and nowadays, it is achieved through artificial intelligence approaches like machine learning and deep learning using various imaging modalities. This study investigates and analyses the various state-of-the-art multi-organ and pancreas segmentation approaches to identify the research gaps and future perspectives for the research community. The objective is achieved by framing the research questions using the PICOC framework and then selecting 140 research articles using a systematic process through the Covidence tool to conclude the answers to the respective questions. The literature search has been conducted on five databases of original studies published from 2003 to 2023. Initially, the literature analysis is presented in terms of publication, and the comparative analysis of the current study is presented with existing review studies. Then, existing studies are analyzed, focusing on semi-automatic and automatic multi-organ segmentation and pancreas segmentation, using various learning methods. Finally, the various critical issues, the research gaps and the future perspectives of segmentation methods based on published evidence are summarized.