Signal Processing, Telecommunication and Embedded Systems: Automation and Sustainability Applications

Remote Sensing (RS) is a technique used to gather and interpret information from a distance using sensors. It provides critical data by capturing and analyzing sensed information. It plays a crucial role in mapping of LULC (land use and land cover) indispensable, for effective planning and supervision. This study employed a deep neural network to perform SS (Semantic Segmentation) on LISS-III multispectral images, using Fully Convolutional Networks (FCN) based on U-Net. The technique of SS (Semantic Segmentation) entails of assigning every image element in an illustration to a specific class. Three innovative datasets were developed for the research (Dataset 1: 1,470 images, Dataset 2: 13,500 images dataset 3: 960 images), consisting of LISS-III images with four color channels: Blue, Green, Red and near-infrared region. This dataset includes False Colour Composite (FCC) and ground reference mask images. The experiment identifies four distinct classes: Aquatic areas, Plant cover, Untilled land, and Urban settlements. The model is leveraged to train input images of size 256 × 256 × 3, 128 × 128 × 3 and produces an output matrix of size 256 × 256 × 4, 128 × 128 × 4 representing a One-hot vector mask for the classes. The experimental results demonstrated that the FCN classifier is highly effective in detecting land-use and land-cover classes, achieving an overall accuracy (OAA) of 81% for Dataset 1, 84% for Dataset 2, and 77% for Dataset 3.

The issue of student assessment becomes even more significant as computer scientist as a subject matter expands its bounds. These assessments are important instruments to evaluate students’ achievements; therefore, the need for a complex question generation system is obvious. It becomes almost impossible for the educational institutes to set the ideal question paper. The traditional approaches are also inconvenient since the questions have to be written by the lecturers. In this case, our system has it easier for the faculty to generate random questions out of the many in the large question database based on the set of chapters made available. This paper focuses on the following research and review techniques used in the development of question papers, particularly using algorithms, databases, and bloom’s taxonomy of six dealing with generation of good examination questions in relation to other learners depending on hierarchies and sections selected by teachers. Through question paper generation, we envision improving the processes of education in a teaching-learning environment.

Mango leaf diseases critically impact both quality and yield in global mango cultivation. Traditional detection methods, reliant on manual inspection, are labor-intensive and error-prone. The proposed model successfully mitigates the gradient vanishing problem, enhancing training stability and improving overall model performance. This work introduces a novel method employing a deep residual skip-connected convolutional neural network (DRCNN) for the identification and categorization of diseases affecting mango leaves. Extensive analysis of the DRCNN, including measures like batch sizes and learning rates, revealed that at 0.001 learning rate and batch size of 8, the network yielded the highest performance. This configuration achieved a remarkable precision of 99.76%, an accuracy of 99.75%, an F1-score of 0.9975, and a recall of 0.9976. Results on the mango leaf dataset demonstrate the proposed architecture’s superior accuracy over traditional and contemporary techniques. By automating disease detection, this framework facilitates timely and precise management, significantly enhancing mango production's productivity and sustainability.

Consumers and businesses face financial risks because of the increase in credit card fraud brought on by the boom in digital transactions. Traditional fraud detection technology can’t keep up with evolving fraud strategies, which leads to high false positives and undetected fraud. This paper proposes a hybrid deep learning system that integrates Autoencoders, Conv1D, SMOTE, and LSTM to increase the accuracy of fraud detection. SMOTE addresses class imbalance, Autoencoders extract complex transaction patterns, Conv1D detects local dependencies, and LSTM captures long-term temporal correlations. Class imbalance is addressed by SMOTE, complicated transaction patterns are extracted by Autoencoders, local dependencies are detected by Conv1D, and long-term temporal correlations are captured by LSTM. When compared to traditional models, experimental results on the European Credit Card Dataset demonstrate improved precision, recall, and F1-score. The results highlight how crucial hybrid deep learning is to create adaptive fraud detection systems that can react to new fraud trends. To improve financial security, future work will concentrate on real-time deployment and enhancing model interpretability.

The coexistence of humans and wildlife has led to an increasing need for early detection and warning systems to mitigate potential dangers posed by dangerous animals. This novel approach addresses this challenge by harnessing recurrent neural networks (RNNs) for real-time detection and alerting. The system utilizes a deep learning framework to continuously analyse environmental data streams, such as live camera feeds from critical wildlife crossing areas and animals perambulate areas. This data source provides a multi-model input stream for the RNN, enabling the model to learn complex patterns associated with dangerous animal presence. When a potential threat is detected, the system triggers alerts through various communication channels such as mobile apps, mails, and sirens. The results offer a promising solution to mitigate the risks associated with wildlife hazards, fostering coexistence between humans and wildlife while safeguarding lives and the natural world.

In our fast-paced, modern world, we tend to overlook the very real and urgent safety problems that women confront when navigating the world outside of their homes because of their homes because of our collective fixation on the glamour of modern living. Unfortunately, the number of kidnappings and dangerous situations has increased, which has negatively impacted women safety. Fortunately, a world of new possibilities has been made possible by the wide spread use of smartphones, which are outfitted with and advanced range of sensors like accelerometers, GPS, magnetometers, multiple microphones, and potent cameras. Sensor networks have easily migrated into many different fields, including as intelligent transportation systems and creative features that were previously unthinkable. Considering previous efforts to improve women’s safety, our proposal centers on creating a novel system that utilizes the underutilized.

This paper presents the design and implementation of a Microcontroller-based Coal Mine Safety Device integrated with advanced Embedded Systems. The primary objective of this innovative device is to mitigate the myriad of hazards and risks inherent in coal mining operations, thereby safeguarding the health and well-being of coal mine workers. Additionally, it prioritizes environmental considerations, ensuring a safe working environment conducive to optimal productivity. Through the strategic utilization of cutting-edge technology, this device offers comprehensive monitoring and response capabilities tailored to the dynamic conditions prevalent within coal mines. By addressing critical safety concerns and environmental factors, this development represents a significant advancement in promoting both occupational health and environmental sustainability within the coal mining industry. In the model, a range of sensors is incorporated to monitor temperature, humidity, and Air Quality Index (AQI) within coal mining environments. Through a meticulously devised plan and algorithm, the operation of both the microcontroller and the sensors is conducted autonomously, with their designated tasks being executed in alignment with the paper’s objectives. Auxiliary to this setup are strategically positioned slave nodes tasked with detecting pertinent environmental data, which is then transmitted to a central master node. This master node, when interfaced with a monitor, enables the remote monitoring of temperature, humidity levels, and Air Quality Index (AQI) without necessitating physical presence in the field.

One of the biggest problems in computer vision and nutritional science is accurately calculating the number of calories in food from pictures. In order to encourage better eating practices, this study tackles the challenge of automatically identifying food items in photos and supplying the relevant nutritional data, especially calorie levels. The YOLO (You Only Look Once) algorithm is used in the suggested system for effective and real-time food item detection. Its high accuracy and speed allow it to recognize several objects in a single image. A pre-existing nutritional dataset is then mapped to the detected food items, providing calorie information based on the meal type and approximate portion size. This method is unusual because it combines YOLO for food detection with a calorie estimating pipeline that is intended to function flawlessly in real-time via an intuitive online interface. This system, in contrast to conventional approaches, provides a fair balance between speed and accuracy, which makes it appropriate for real-time applications in diet planning, exercise, and healthcare. According to experimental results, the system is resilient in identifying a variety of foods and providing accurate calorie estimates, while also significantly improving processing efficiency when compared to traditional approaches. This study opens the door for further advancements in food recognition and dietary analysis while offering insightful information about developing automated nutritional evaluation systems.

Breast cancer is one of the leading type of cancer impacting women globally, early diagnosis of such diseases helps in identifying prevalent symptoms in the early stages hence this study focuses on choosing best machine learning approaches to tackle this problem. In this study, machine learning approaches are presented to classify breast cancer sub-types based on high dimensional Gene Expression data. 151 samples with 54,676 gene activity features per sample have been obtained from Kaggle which are clustered into Basal, HER2-enriched, Luminal-A, Luminal-B, Cell Line, and Normal. The data set was rectified for class imbalance by the means of the Synthetic-Minority-Oversampling Technique (SMOTE). To enhance performance of model and to reduce the computation power needed, the feature selection was done using advanced optimization methods such as Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Multiple Metrics like Accuracy, Positive Predictive Value, Sensitivity, and F1Score were executed to assess the performance of various machine learning algorithms including Support Vector Machine (SVM), Random Forest (RF) and Artificial Neural Networks (ANN). The results proved that the ANN model had maximum accuracy of 98% after feature selection. This demonstrates the success of these techniques in boosting classification accuracy. Using micro-array, this study seeks out the opportunities of machine-learning enhancement in breast cancer examination.

In the field of agriculture, choosing a right crop for a particular land is very important. For such recommendation, we need soil nutrients data and environmental data. Also, predicting the post-harvest price of the crop at the preparation stage helps the farmer to harvest the crop, which gives him profit. To recommend the three crops which can be cultivated on the land, Stacking Classifier with base models as K-Nearest Neighbors, Naive Bayes, Decision Tree and Support Vector Machine is used. The stacking classifier model is a machine learning ensemble which combines multiple base models to enhance accuracy. For these recommended three crops, the prices are predicted using LSTM model. The LSTM model is particularly useful in time-series forecasting. The accuracy of the crop recommendation model is 96.61% and Mean Absolute Percentage Error for the price prediction model is 5.6%. The system has the potential to increase agricultural production and profitability.

While e-commerce platforms are growing rapidly across category verticals, from clothes, groceries, and household products, digital literacy has also emerged as a major contributor to efficiently maneuvering through shopping experiences via portals. All of these platforms can be very helpful, but they are also very complicated, and people who are not familiarized with digital technology under the usual intersects may find them difficult to use. When users are tasked with long or complicated shopping lists, the complex search functionality mixed with product categories and filters becomes a blocker for a substantial segment of users. However, the process of entering each toe to a cart is sloppy and tedious, which may annoy people who are just shopping, trying to eat a fucking steak as well as fucking potato. Not only this makes it less welcoming to the less tech-savvy but it results in a more cumbersome experience for those who simply want to shorten their time on the e-commerce portals. This project suggests a way through using its optical character recognition (OCR) technology by means of the open-source Tesseract engine, in an effort to facilitate the online shopping process. It is used to scan an image of a handwritten shopping list and extract the associated text of the items so each individual item on the list could be recognized and added to the cart automatically by the system rather than manually inputted which consumes a longer time. It makes the process quicker and easier as your search each item. This resolution conveys not just robustness in time and effort used in the present shopping event yet, in addition, makes the shopping understanding increasingly open to clients who may need computerized ability. This method is suitable to a mix of needs and abilities by providing a seamless alternative for those who might find online purchasing through an online retailer interface daunting as well as to individuals who might be pushed for time. This approach aims to reduce barriers to online shopping and enhance accessibility, making digital platforms more usable, inclusive, and appealing for all, regardless of technological experience.

Low-frequency antennas below 1 GHz are dragging the attention of researchers due to its applications such as defense, mining, and underground communications. Designing a miniaturized structure with acceptable performance is a cumbersome task. As the size of the patch antenna increases with decrease in the frequency, appropriate techniques have to be adopted to design a patch antenna at low frequency. This paper presents a dual band transverse electric (TE) mode exited antenna that works well below 1 GHz. The principle of dual periodic artificial magnetic conductor (AMC) is the key for the dual band in this proposed antenna. The proposed antenna resonates at 515 MHz and 775 MHz with gain of 2 dBi and 4.5 dBi, respectively.

These days, researchers are designing adaptive controllers for a range of applications because of their improved robustness and performance. Nevertheless, for under-actuated inverted pendulums, traditional adaptive controllers have not demonstrated performance that is sufficient. Therefore, in order to solve the aforementioned issue, a unique optimal fractional-order Massachusetts Institute of Technology (MIT) and Lyapunov rule are created for direct model reference adaptive control (MRAC) scheme in this work. To avoid time-domain analysis, an indirect method for higher-order continuing fraction expansions approximation is used to solve the fractional portion. Using a multi-objective function, the gray wolf optimizer (GWO) determines the changeable parameters of the aforementioned MRAC schemes (fractional-order parameter and adaptive gain). In simulation, comparisons are conducted between the FOPID controller, integer-order Lyapunov (IOLY)-based MRAC, fractional-order Lyapunov (FOLY), FOMIT rule-based MRAC, and integer-order MIT (IOMIT) rule-based MRAC schemes. A comparison with previous research is also displayed. To demonstrate the robustness of the proposed control technique, the effect of perturbing the plant model is examined. To demonstrate the effectiveness of the proposed approach, the effects of measurement noise and nonlinearity in plant dynamics are further investigated. The optimal FOLY approach that has been proposed performs better than the other schemes, as confirmed by the results.

Diabetes affects millions of individuals, making it a global health issue, and is categorized as a chronic metabolic disorder. The driving purpose of this study is to develop a predictive model for diabetes based on machine learning algorithms so that early assessment of and intervention for diabetes risk can be done. Diabetes risk assessment and early detection are components of the preventive healthcare approach; the main aim is to assess and compare the performance incorporating diverse machine learning algorithms for accurate diabetes risk prediction, and each method examines the implementation and refinement of hyperparameters for best prediction performance. Such indicators include precision, accuracy, recall, F1-score, and the receiver operating characteristic (ROC) curve, which assesses the area that the curve encloses as one of the evaluation metrics. Methods such as gradient boost, XGBoost, random forest, and lightweight GBM are notably helpful since they can combine the benefits of many models to improve prediction accuracy, and the final goal is to formulate reliable and accurate predictive models. Certain evaluations are made to evaluate the effectiveness of these models, and the results of the study are necessary to create preventive medical care and adjustment of treatment of patients. This study aims to develop new health-improving methods and make a considerable impact on the prevention of diabetes, utilizing ML algorithms.

Speech and written words are examples of natural language used by humans to communicate with one another. Nonetheless, the only way to communicate for the deaf is through sign language. They cannot converge with one another in the absence of an interpreter. For this reason, the development of technology that can read sign language will greatly benefit the social lives of the deaf. Unlike American Sign Language (ASL), Indian Sign Language (ISL) employs both hands for movements rather than just one. Creating an ISL recognition system that translates sign language into legible text is the recommended course of action. There are several ways to accomplish this, including object and position detection, smart gloves, and more. The suggested technique detects hand moments by using TensorFlow object detection. To extract features, MobileNet is utilized in conjunction with single-shot multibox detection (SSD) for further detection.

Diabetes is a chronic condition, and with the increasing prevalence of cases globally, there is a heightened need for innovative IT solutions that facilitate early detection and effective management to minimize its impact on individuals and healthcare systems. This research evaluates the performance of two machine learning models–Random Forest and XGBoost in predicting the onset of diabetes. The Random Forest algorithm achieved an impressive 95% accuracy, outperforming XGBoost, which reached 91%. Moreover, by utilizing the Apache Spark framework, we reduced the model’s training time to just 5–10 seconds, a considerable improvement compared to the traditional Random Forest model, which required 16 min. To put these results into practical use, we developed a web application utilizing the Apache Spark Random Forest model to assess diabetes risk in new cases, offering personalized health insights. This study highlights the potential of machine learning for early diabetes prediction, demonstrating the superiority of the Random Forest model and suggesting areas for improving XGBoost. Overall, it showcases how predictive modeling can be leveraged for managing chronic conditions and advancing personalized healthcare solutions for diabetes.

The goal of this project is to detect anomalies in financial market data using an LSTM Autoencoder model created with TensorFlow Keras API. The dataset is made up of daily index values from multiple decades and is divided into two sets: training (80%) and testing (20%). The LSTM Autoencoder is a form of RNN that uses an encoder–decoder architecture to replicate input sequences and identify common patterns in data. Anomalies are discovered by calculating the reconstruction error, which is the difference between the original and rebuilt sequence. A threshold derived from the training error distribution indicates significant deviations from normal data. A threshold derived from the training error distribution identifies large departures from normal data. Data points with reconstruction errors that exceed this threshold are classified as anomalies. The capacity of the model to distinguish between normal and anomalous data items is evaluated using precision, recall, and F1-score. Visualizing training and validation loss, as well as highlighting anomalies, reveals unexpected behavior.

An inventive way to improve waste collection and recycling is to use barcodes on smart waste containers. These containers have barcode scanners that make it possible to efficiently identify and group waste according to its type and destination. By scanning the barcodes on waste containers, the system can automatically sort and monitor the waste, reducing the need for manual labor and increasing the accuracy of the waste management processes. Additionally, by using the data obtained by these smart containers to identify areas where efforts to lessen waste and improve recycling may be enhanced, a more sustainable and environmentally friendly method of managing waste may be attained. Overall, using smart containers for waste with barcodes is a technology that has promise for improving waste management in cities and municipalities while minimizing environmental effects.

The transformation of handwritten notes into digital text is a valuable tool across various industries such as education, healthcare, and business. Our research aims to develop a system that utilizes Convolutional Recurrent Neural Networks (CRNNs) for effective and precise handwriting recognition. The system’s objective is to accurately convert handwritten content from images into digital, editable text. To achieve this, a deep learning model powered by CNN is employed to interpret handwritten text, identifying key features and structures within the images. Convolutional Recurrent Neural Networks (CRNNs) are preferred for their ability to identify complex patterns within visual data, making them well-suited for tasks such as recognizing various handwriting styles. The system strives to provide a high level of accuracy in text conversion, ensuring minimal errors and enhanced user experience. The findings from this project illustrate the potential of CNN-based models to streamline the process of digitizing handwritten materials, minimizing the need for manual intervention and enhancing overall efficiency. This paper provides an overview of the techniques used, the challenges faced, and the results achieved highlighting the effectiveness of modern deep learning methods in solving practical problems in handwriting recognition.

The rapid increase in vehicles because of urbanization has created chaotic transportation networks causing an excessive delay in travel, consumption of fuel, and air pollution, becoming one of the major issues across the globe in urban places. Therefore, affecting mobility, safety, and quality of life. Addressing this pressing issue, this paper develops a dual ensemble regressor model and for predicting traffic management based on machine learning in order to enhance traffic predicting accuracy. Furthermore, our paper compares the results of ensemble model with the other widely used regression models including Random Forest, CatBoost, Gradient Boosting, K-Nearest Neighbours, Ada Boost, Support Vector Regression, Linear Regression. The models have been employed to a Kaggle dataset containing 33,707 features and 15 instances. This paper also underscores the past research literatures that have approached this serious issue. The results show the superiority of ensemble modelling approach greatly improving the predictability of traffic volume, yielding high accuracy metrics including low Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and high R-squared values. The proposed model gives the RMSE and MAE as low as 342.50 and 205.47, respectively, and an impressive R-squared value of 97%, greater than all the other models compared. This work aims to add to the existing knowledge body on traffic prediction and traffic management authorities, to overcome congestion and promote efficiency within transportation systems, while helping to mitigate negative environmental impact.

Time series analysis plays a vital role in extracting meaningful information from temporal data. Forecasting Time series data is a powerful tool for businesses to make informed decisions, improve efficiency, and stay ahead of the competition. This study aims to fit an appropriate model for the Air traffic data. The data series exhibited cyclical patterns and/or seasonality. Both of these properties were taken into account in the time series models utilized in this study. In the analysis of this data, both smoothing methods and ARIMA models were used. Since seasonality is a complicated factor in a time series, findings inferred that SARIMA models perform well. Based on the accuracy measures we could infer that the models offer a promising means of forecasting air passenger demand.

In our daily lives, music plays a vital role in soothing the emotions of its listeners. They can experience a distinct emotional reaction. We can tell someone’s emotion just by looking at their face. By directly identifying the user's facial emotions, the current work aims to create an automated system that creates playlists and plays music based on the user's mood. This eliminates the tiresome and repetitive process of physically compiling songs and assists the users to get automatically generated music playlists that suits their present state of mind. To achieve this, the open-source computer vision library (OpenCV) has been used to identify the user’s emotion. This model needs a camera to record the user's face. Based on the user’s identified “emotion” from the face, the playlist is suggested to them. The overall accuracy of the system is found to be 92%.

This paper analyzes the tendency of adaptation of emerging technologies in schooling education system between the Indian private, central government, and government schools and investigates major factor(s) responsible for poor performance in the government school result. For collecting information on teaching–learning process, a sample was made up of 180 teachers randomly selected from each category of schools either central, private, or government school. Data were analyzed using simple percentages, mean percent, SD and variance, and Z-test with the help of charts using Python libraries. Result analysis has been done on the basis of interview, which has been conducted among teachers and administrators of central, private, and government schools. Results showed that pupils in the private and central schools performed better than the government schools, and adaptation of technology is rising in private school comparatively. Private and central schools’ management system, teaching–learning process, and evaluation procedures are found to be better than government schools.

Machine learning is a sub-area of artificial intelligence. It finds its relevance significantly in changing daily routines, interactions, experiences, and decision-making in different fields, for instance, health, lifestyle, agriculture, and finance. This research explored the application of machine learning-based algorithms to identify stress from social media posts with an emphasis on text classification through the platform Reddit. Experimental methodology: A study of employment algorithms like Naive Bayes and Support Vector Machine (SVM) alongside data preprocessing and feature extraction techniques such as Bag-of-Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF). The results indicate Naive Bayes to have a higher accuracy than SVM which was at 67% while the former at 73.9%, demonstrating the possibility of employing models for automated stress detection. Such findings stress the importance of machine learning when it comes to mental health monitoring applications and the use of social media for identifying and addressing stress-related problems.

Wearable devices are solution for real-time stress and health monitoring, leveraging, and body temperature. Of current, technologies reveal diverse applications, from simple consumer electronics to sophisticated biomedical systems, designed to enhance mental health management. These devices utilize advanced algorithms to detect stress levels, providing users with actionable feedback to mitigate anxiety and improve overall well-being. Challenges remain, including the need for improved accuracy, user comfort, and data privacy. Future developments in wearable technology promise to integrate more comprehensive health metrics, paving the way for personalized health interventions and enhancing the efficacy of stress management strategies in everyday life. The increasing prevalence of stress-related health issues has driven the development of wearable devices that monitor physiological indicators for real-time stress detection and health management. This comprehensive review explores various wearable technologies, including smartwatches, fitness trackers, and specialized biomedical sensors, which utilize metrics such as heart rate variability, skin temperature, and electrodermal activity to assess stress levels. Recent advancements have led to the integration of sophisticated algorithms that analyze these signals, offering users actionable insights to manage their stress effectively. Despite the promising capabilities of these devices, challenges such as accuracy, comfort, and data security persist. Moreover, user engagement and adherence to feedback remain critical for the success of these systems. Future research should focus on enhancing the reliability of measurements, expanding the range of physiological parameters monitored, and ensuring seamless integration with mental health applications. The ongoing evolution of wearable technology holds significant potential for personalized health interventions, ultimately contributing to improved mental well-being and resilience in an increasingly stressful world. Wearable devices for stress and health monitoring have emerged as transformative tools in personal healthcare, leveraging advanced technologies to provide real-time insights into physiological responses associated with stress. These devices utilize various sensors to continuously capture critical physiological signals such as heart rate variability, electrodermal activity, skin temperature, and respiration, enabling users to monitor their stress levels and overall health more effectively. The integration of sophisticated algorithms and machine learning techniques enhances the accuracy of stress detection, allowing for personalized feedback and interventions tailored to individual physiological profiles. Recent studies highlight the effectiveness of these wearables in correlating physiological changes with self-reported stress, validating their utility in real-world applications. User satisfaction is notably high, with many individuals reporting an increased sense of control over their stress management through the continuous monitoring and actionable insights provided by these technologies. Despite the advancements, challenges such as ensuring data privacy, enhancing sensor accuracy, and improving user comfort remain critical areas of focus for future research and development. The ongoing evolution of these devices points toward their broader application in mental health management, fostering a proactive approach to healthcare that encourages individuals to engage in preventive measures. By empowering users with real-time data and personalized recommendations, wearable technologies are poised to revolutionize how individuals interact with their health, ultimately contributing to improved mental well-being and quality of life. As research in this field progresses, it is anticipated that these innovations will not only enhance the efficacy of stress monitoring but also integrate seamlessly into everyday life, promoting healthier behaviors and a deeper understanding of the intricate relationship between stress and overall health.

Pneumonia is a major global health issue, with prompt diagnosis critical for effective treatment. Chest X-rays (CXRs) are an important diagnostic tool, but manual interpretation can be difficult and error-prone. This paper offers a feature-level ensemble learning strategy for automating pneumonia diagnosis in CXRs, which employs convolutional neural networks (CNNs) to improve diagnostic accuracy and assist healthcare practitioners. The model combines features from three pre-trained CNN architectures—VGG19, EfficientNet-B0, and DenseNet121—taking advantage of each architecture’s specific strengths. By adding trainable weights to the features collected by multiple models, instead of only concatenating features, this method enables the model to dynamically prioritize the most relevant feature representations while capturing a greater variety of features. In order to improve interpretability and assist physicians in making decisions, gradient-weighted class activation mapping, or Grad-CAM, is also utilized to produce heatmaps that highlight important regions in CXRs. The proposed model obtains 93.1% classification accuracy on the test dataset, demonstrating its ability to aid in the precise and efficient detection of pneumonia. This study deals with the need for dependable AI tools in medical diagnostics, with the goal of reducing radiologist burden and improving decision-making in resource-constrained environments.

Multilevel inverters play a pivotal role in modern power electronics, offering enhanced performance in terms of efficiency and voltage quality. This paper presents a novel 35-level asymmetrical multilevel inverter utilizing 12 switches and 6 DC sources (where two Vdc, two 3 Vdc, and two 5 Vdc are used). In renewable energy-based systems such as photo voltaic cells, or hybrid electric vehicles, several modules are connected to attain increased voltage levels. The proposed topology can be used in such applications where multiple DC sources are used. By cascading the same configuration of the proposed topology, a higher number of voltage levels can be achieved with good quality output voltage with lower THD. Simulations demonstrate its effectiveness along with voltage stress analysis of the topology. Comparative analysis is done with existing 35-level topologies concerning device count, source diversity factor, switch diversity factor, and many sources required. The circuit efficiency, switching losses, and conduction losses are found to be nominal for their use in hybrid electric vehicles.

In a data-driven world, every type of data requires a physical drive in order to be stored. The number of photos, videos, and other comparable data kinds that are shared online has increased dramatically. Efficient storage and transmission of data has made image compression an important topic of research. This paper proposes two recent deep learning-based image compression frameworks to get encouraging advancements and better rate-distortion efficiency when compared to classical or traditional image compression schemes. Autoencoders and vision transformers eliminate the limitations of fixed mathematical models because they deal with more advanced representations of image content, capturing complex patterns and global properties of an image. This paper work has empirically investigated the impact of two deep learning-based frameworks based on image compression and also, we have compared them by evaluating their mean square error, peak signal-to-noise ratio, and compression ratio using CIFAR-10 datasets to find out which deep learning approach gives good promising results maintaining image quality to expand the scope of image compression. Experimental results reveal that autoencoders, with their convolutional structure, provide a balanced approach in reducing image dimensions and also preserves a reasonable image quality. Further, they yield reasonably good peak signal-to-noise ratio (PSNR), low mean square errors (MSE), and high ratio of compression which make them relatively efficient for compression. In addition, they require low computational cost and complexity as compared to other methods. On the other hand, vision transformers (ViT) yield very high peak signal-to-noise ratio (PSNR) values but due to its aggressive compression it degrades the image quality. Vision transformers are comparatively expensive computationally and also its complex training architecture may not make it resource friendly.

To examine increasing concerns regarding digital dependence, this study uses machine learning techniques to measure students’ dependency on electronic devices. As academic and social activities gradually depend more on smartphones, computers, and tablets, the potential risks posed on one’s mental health and academic performance become much greater. The study analyzes multiple machine learning algorithms to calculate the trends for factors like screen time, usage behaviors, and other psychological variables that contribute to dependence on devices. Using surveys, along with usage analytics, makes it possible to construct a model that will predict the likelihood of addiction toward the analyzed devices. This obsession with technology serves as vital indicators for preventing its overindulgence, preparing educators and parents for the initiation of healthy digital habits. It also shall be beginning the tide toward more comprehensive talks, thus adding to the understanding of the phenomena of technological dependence on how best to achieve the balance and safety enhancements of digitization in education. These are steps that you can implement to minimize the risks of developing such dependency, while still using technology within an ecosystem of learning and communication.

Load balancing is a challenging aspect on the increase in the performance and improvement of the utilization in the grid environment. Prediction of future load is a prime factor in balancing of load. Many attempts were made in the past on prediction of load with simple classifier approaches. There are many mathematical and machine learning models available on the numeric predictions of the parameters on the available data. The objective is to anticipate the demand of the load in the grid using numeric prediction methods with machine learning models. In this research, linear regression, Gaussian process and generalized linear model are suggested among many mathematical and machine learning models on the prediction of the load against the standard data set with respect to the suitability and close association of the parameters of load. A comparison is made on the performance of the prediction of the load among these selected models.

Land cover classification plays a critical role in environmental monitoring, agriculture, urban planning, and disaster management. Traditional methods for land cover classification from satellite imagery are often labor-intensive and prone to errors, highlighting the need for automated and efficient solutions. This project investigates the use of deep learning methods, particularly Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), for classifying land cover in high-resolution satellite images. The objectives include developing and evaluating deep learning methods to automate the land cover classification task and improve classification accuracy compared to traditional methods. Experimental results demonstrate promising performance, with CNNs achieving an accuracy of 86.12% and ViTs achieving an accuracy of 87.39%. The models exhibit competitive performance in terms of accuracy, efficiency, and scalability, highlighting their utility for any real-world applications in environmental monitoring and geospatial analysis. Future research directions include exploring multi-modal data fusion techniques, advanced attention mechanisms, and integration of domain-specific knowledge to further enhance the capabilities of deep learning methods for classification of land cover.

The study focuses on the design and performance evaluation of a passive magnetic bearing system for horizontal rotation, utilizing repulsive magnetic technology. This involves designing and analyzing the performance of the permanent magnet (PM) bearing system intended to rotate on a horizontal axis. The proposed PM bearing is based on passive magnetic technology, and its performance is evaluated by using both analytical solution and finite element analysis.

Putting harmful brain activity into groups has become an important area of study that has a big effect on finding and stopping neurological disorders (ND). Because about 3 billion people around the world have some kind of ND, and it can take long time and hard to figure out what’s wrong with them. On average, they have to wait two years for a final diagnosis. It takes a long time and a lot of different tests to confirm that someone has an ND. Some of these tests are non-invasive, like EEGs, ultrasounds or simple reflex tests. Others, like spine taps or biopsies, are more invasive. Electroencephalogram (EEG) is one of the first tests that is usually done on a patient. It is done by putting several devices on the head and measuring the electrical activity of the brain. This shows patterns of neural activity happening inside the brain. It is hard to decode brain patterns or put them into groups. So, the goal of this article is to shorten the time to diagnose using a long short-term memory (LSTM) model to find harmful brain activity patterns in spectrogram data. This could lead to better outcomes for patients.

Following investigations on youth drug use, the teenage population’s substance use is attributable to a number of social, psychological, and environmental causes due to its generation from the root of agriculture. Rehm et al. (Addiction 98:1051–1060, 2003 [1]) emphasize the need to understand teenage drinking cultures in their quest to understand the relationship between these patterns of drinking and the consequences that follow. While Brook et al. (Addict Behav 26:103–116, 2001 [2]) do find that certain cultures affect teenage patterns of marijuana use, family, and socioeconomic status are more decisive factors even with respect to such patterns of substance usage. Pro-drug norms affecting juveniles are the chief thematic concern of Sussman et al. (Subst Use Misuse 37:111–123, 2002 [3]), who demonstrate how such distortions in perspective lead to unprecedented rates of substance abuse among the most at-risk youth populations. Risk factors are also identified by Chen and Kandel (Am J Public Health 85:41–47, 1995 [4]), who provide a longitudinal overview of how drug use order evolves at different life transitions between adolescence and mid-thirties. At the same time, Lundborg (J Health Econ 24:279–299, 2005 [5]) studies the role of economic determinants in the aggressive consumption of alcohol while Hawkins et al. (Psychol Bull 112:64–105, 1992 [6]) review risk and protective factors associated with the prevention of substance abuse. The social learning theory as it pertains to substance abuse and academic achievement is explored by Akers (Social learning and social structure: a general theory of crime and deviance. Northeastern University Press, 1998 [7]), and the connection between drugs and education by Van Gundy (Sociol Spectr 26:117–142, 2006 [8]). Finally, a comprehensive account of drug and alcohol consumption among students in European countries is presented by Hibell et al. (The 2011 ESPAD report: substance use among students in 36 European countries.). The Swedish Council for Information on Alcohol and Other Drugs, 2012 [9]) and their associates with emphasis on cross-border incidence and determinants.

Artificial Intelligence (AI) refers to the ability of machines to think intelligently like human beings. Augmented reality (AR) involves overlaying virtual objects on physical objects to enhance the experience of reality. AR technology necessitates various types of triggering, including marker-based, geographic or location-based, marker less, and position-based. This paper seeks to understand how young students pursuing higher education across India perceive AI and AR in the education sector. This study contributes to finding the factors influencing the adoption of AI in the education system. We sourced the literature from the Google Scholar and Scopus databases, using English-language criteria such as adoption and perceptions of AI, AR, and technology. We conducted an online survey involving approximately 200 students from all over India. We conducted this analysis in Excel. This study shows that around 90% of our young generation are familiar with the latest technologies. An interesting finding from this study is that over 65% of students use AI technology for educational purposes. Researchers have also observed significant barriers to AI integration, including data security and privacy concerns, lack of resources, biased results, cost of implementation, resistance to change, and lack of regular training for teachers. Addressing these challenges is crucial to increasing AI adoption and maximizing its potential in education. Most students believe AI in education could lead to biased predictions of academic performance, as algorithms are trained on historical data that may reflect inherent biases. Hence there is a need to train these algorithms. Around 72% of youths believe that AI-based monitoring tools record excessive amounts of their activities. AI monitoring tools should design clear limits on the type and scope of recorded data to address this issue and ensure user privacy. We must establish transparent policies and user consent protocols to enable students to manage their data while utilizing AI-based tools. Young people also believe that integrating AR with AI will be beneficial and enhance their learning experience. Future researchers should further analyze the factors influencing the adoption of AI in education using qualitative and quantitative methods with more theories and diverse samples, respectively.

The restaurant industry is susceptible to discrimination that affects customers as well as staff. This can be manifested as discrimination on a racial or gender basis, class or other factors which affect the way patrons are treated, the services offered, and employment opportunities. Studies note that races or ethnic minorities are usually underserviced, occupy inappropriate tables at best and some sources highlight that gender may be a discriminatory factor in hiring and promotions in the cash on the staff’s hands. People also are treated differently depending on their class where people with lower income may not be normed into hiding certain facilities in the restaurant. These forms of discrimination are harmful great concentration of local authorities and community members are needed to change that. Among possible responses to these issues the authors view the distribution of an in-depth bias training, quotas for those of a different race or gender to be employed, and clearly defined processes so the customers and the staff are equally honored and respected.

An innovative traffic sign detector (TSD) system has been developed. Using a Convolutional Neural Network (CNN) trained on a large dataset of annotated traffic sign images, this system combines state of the art technology. The ability to recognize, classify, and localize traffic signs in images and videos allows for better decision making in road safety and traffic control matters. Its adaptive learning mechanisms and domain adaptation strategies allow generalization of the system in various environmental conditions ensuring consistent performance and reliability in real-time scenarios. The role in road safety and traffic management is immense: increasing awareness and compliance of drivers, supporting intelligent traffic management systems to reduce accidents and improve traffic flow. In parallel, autonomous technologies and V2X communication combine to refine traffic flow dynamics. Work on robust theoretical models and novel algorithmic frameworks for autonomous traffic management. The idea is to combine this with next generation V2X communication to create a vehicle—infrastructure communication environment where vehicles communicate with each other and with the infrastructure to improve traffic coordination, congestion reduction and road safety. Extreme simulation-based analyses of proposed systems in dynamic traffic scenarios are performed to evaluate their adaptability and performance in terms of real-time data exchange, predictive analytics and unforeseen events. Ethical questions, data privacy questions, and broader social implications form part of this research. With cars communicating with vehicles and their surroundings, ethical and privacy concerns must be addressed for a responsible and secure implementation of these technologies. It offers insights into the current discourse regarding intelligent traffic control systems aiming at driving the design and implementation of sophisticated traffic management solutions for a safer, more optimized, and ethically sound urban mobility ecosystem. The ultimate vision is a future in which autonomous technologies and V2X communication reshape urban transport dynamics.

Wild animal invasion has been on the rise in agricultural fields, causing significant damage to crops. Hence, a solution is proposed that uses neural training with the backpropagation algorithm to identify different animals based on their forms, surface details, and color. This training process contains multiple perspectives of wild animals—frontal, back, and lateral views—focusing on features like coloration, form, and patterns. A live camera feed is examined through the MATLAB video processing algorithm to extract image features and compare them with a real-time neural-trained database. The database is created with wild animal image features along with their disturbing frequency range. The similarity result of the database with the camera feed will trigger a speaker with a stored frequency level. Thus, the novel idea of this work is to produce a sound signal that can alert humans about the presence of animals and prevent the animals from entering the fields.

The burgeoning growth of data in recent years has led to datasets becoming increasingly larger, which poses challenges for the development of machine learning models. The complexity of contemporary machine learning algorithms often results in slower training times and longer development periods. To address these issues, this paper proposes a novel numerosity reduction algorithm that leverages the Central Limit Theorem (CLT) to reduce the number of data points for training machine learning models while preserving interpretability and conforming to a Gaussian distribution. The proposed numerosity reduction algorithm is expected to enhance the training time of binary classification models and improve the performance of algorithms that rely on the assumption that the training features are Gaussian distributed. Consequently, this technique may facilitate the development of large and complex machine learning algorithms, making them more feasible to train and deploy. The aim is to decrease the number of data points (cloud points) used for training machine learning models, while preserving interpretability and transforming the distribution of the variables to Gaussian. This is particularly advantageous as several algorithms assume data to be normally distributed. By harnessing the properties of CLT, it is expected that training times for binary classification algorithms will be enhanced, as well as the performance of algorithms that rely on gradient descent for optimization. This speed in training is pivotal for the development of large and complex machine learning algorithms.

This approach provides network efficiency in distributed manner of configurated environment without loss of minimal level data transmission. In future, it may be extended to neural and fuzzy networks. The proposed system provides the specific outputs related to maximum degree based node of the given clustered network. This approach leads the new way of finding shortest weighted linear path with N points.

Extended-Phase-Shift (EPS) mode offers a promising approach to improving the efficiency of isolated bidirectional DC–DC converters compared to conventional single-phase-shift control methods. This research paper investigates the dynamics of power flow concerning phase shift ratios and identifies optimal conditions for maximum power transfer. Additionally, it provides a comprehensive analysis of switching characteristics under the extended phase shift operation of a Dual Active Bridge (DAB) DC–DC converter. The EPS control technique introduces two degrees of freedom: an internal phase shift within a single H-bridge and an external phase shift ratio. This enables precise control over power flow and voltage regulation under various load conditions, thereby enhancing converter performance. To control the output voltage and achieve higher efficiency, a Proportional-Integral (PI) controller has been implemented. The converter model operating in EPS mode has been simulated using MATLAB SIMULINK.

Solar energy generation is often hindered by various surface conditions on solar panels, such as dust, snow, bird droppings, and physical or electrical damage, which reduce efficiency. The growing reliance on renewable energy sources emphasizes the need for effective maintenance solutions to sustain optimal performance. This research aims to address the gap in automated fault detection for solar panels by developing an IoT-driven deep learning model to accurately identify and classify these faults. Utilizing MobileNetV3 architecture, combined with data augmentation techniques, the model is trained to detect six fault types, demonstrating robust accuracy in distinguishing between conditions. The innovative use of data augmentation, transfer learning, and MobileNetV3 contributes to the model's enhanced performance and precision, providing significant advancements over traditional manual inspections. Experimental results reveal a classification accuracy of 94%, confirming the model's effectiveness in real-time fault detection. These findings underscore the model’s contribution to optimizing solar energy efficiency and maintenance cost reduction, marking a meaningful impact on renewable energy management.

The rapid advancement of the Internet of Things (IoT) has transformed health monitoring by enabling continuous, real-time data collection, especially for tracking sleep quality and overall health. This research aims to address the limitations of conventional models by developing a robust framework for predicting health and sleep quality using stacked ensemble regressors. The study fills a gap in knowledge by focusing on integrating multiple models to enhance prediction accuracy in a highly dynamic and complex dataset generated by IoT devices. To achieve this, an ensemble approach was employed, combining Gradient Boosting Regressor with other predictive models in a stacked architecture. Feature engineering was applied to develop interaction terms, optimizing the model with hyperparameter tuning. Novel elements of the work include the use of feature interactions, subsampling, and shrinkage, which collectively improve model robustness and generalization. The algorithms contributing to the model's novelty include Gradient Boosting for capturing complex dependencies and stacking methods for refining predictions. Results demonstrate that the proposed model outperforms traditional machine learning approaches in predicting sleep quality and health metrics, as evidenced by significant improvements in R2, mean squared error (MSE), and mean absolute error (MAE). These findings underscore the effectiveness of ensemble methods for handling multidimensional IoT data and provide a pathway for real-time, accurate health insights. The implications are significant for preventive healthcare, enabling more personalized and reliable health recommendations based on real-time data.

Many road accidents now happen because of drivers not paying enough attention or being vigilant. We call this driver sleepiness. This results in numerous unfavorable circumstances that negatively impact people’s life. The driver fatigue identification and appropriate handling of such information is the primary objective of this study. There are numerous techniques that rely on how the car moves or how the driver behaves. The physiological method is one of the techniques that helps keep the driver awake and distracted from his tiredness. A small number of techniques also deal with large amounts of data and costly sensors. As a result, this research creates a mechanism that can precisely and properly identify sleepiness in real time. This technology captures and records the driver's facial expressions using a while simultaneously keeping an eye on the driver's health. Each frame's movements are all recognized. The MOR and eye aspect ratio are calculated. The estimated values are contrasted with the basis given by the module, and the variation in value initiates the detection process. In order to increase efficiency, the vehicle speed will be lowered anytime drowsiness is detected. Additionally, offline machine-learning techniques are used.

Current research is aimed at providing technology that will improve the limb functions of individuals with limb deficiencies. Its focus is on the design, development, and use of state-of-the-art prosthetic hands that are controlled by muscle signals (specifically myoelectric signals) from the user's residual limb. With the use of electromyography (sEMG) sensors, this new concept hand transforms the muscle into a sense, providing the user with a natural and responsive movement experience. The report discusses the integration of biomechanics, signal processing, and robotics to achieve artificial and flexible devices. By combining these disciplines, the model not only restores limb function but also provides mobility and efficiency in use. Key points of this advancement include the integration of in situ electromyography sensors to capture and interpret electromyographic signals, the highest performance signal converting this into useful commands for prosthetic hands, and robotic engineering to convert these commands into signals clear, natural movement.

Individuals with developmental and sensory impairments often face unique challenges that hinder their ability to communicate and perceive their surroundings effectively. One approach focuses on developing effective communication strategies for individuals with limited verbal abilities, such as those on the autism spectrum, by utilizing a structured, visually driven method that encourages users to express their needs and thoughts through non-verbal cues. This method has been shown to significantly improve expressive language, social interactions, and the ability to convey complex ideas, making it a valuable tool in educational and therapeutic contexts. Another solution addresses sensory challenges experienced by visually impaired individuals by providing a device that translates visual information into auditory feedback, enabling users to identify and differentiate colors in real time. This technology empowers users to navigate their environment independently, making informed decisions in activities like clothing selection and object recognition. Both strategies contribute to fostering autonomy, confidence, and social inclusion, demonstrating the importance of tailored assistive technologies that bridge communication and sensory gaps.

Telecom providers are known to offer value-added services as a subscription add-on to the standard core services. In the competitive environment today, a telecommunication company feels that the best approach to measure and improve the business is to be as data-driven as possible. This research aims to develop a model to automate the prediction of average revenue income per user (ARPU). Work done in this paper uses information from the sample business databases which yield about 114,080 rows of data. Several models were built and studied to compare their performances. Four important target variables were considered to appraise the performance of the models. The paper presents two potential strategies to raise the model acceptance rate and considerably increase the model’s accuracy.

The escalating pace of technological advancement worldwide precipitates an exponential demand for experts in the field of computer science. In this dynamic landscape, the requisite skills and proficiency in computer science have become paramount while human resource specialists and recruiters are confronted with an unprecedented amount of Curriculum Vitae (CV) daily, emphasizing the need for a technological solution such as a resume parser to streamline the evaluation process. This study is carried out with the aim of facilitating a more streamlined recruiting process using a text mining-assisted resume interpreter, particularly for Software Engineering positions. With the use of a dataset compiled from a combination of Software Developer vacancies, programming-centered resumes, and non-technical curriculum vitae, a linear Support Vector Machine (SVM) model is constructed and deployed on a web application. To evaluate the model's performance, metrics such as model accuracy, precision, recall, F1, and cross-validation scores were utilized.

The study endeavors to examine the ways in which small and medium-sized businesses (SMEs) adopt sustainable practices (ASP) in technology projects in relation to technical clarity (TC), environmental impact (EI), and organizational culture (OC). It looks at stakeholder commitment (SC) as a possible mediator in these connections. Using a survey provided to key stakeholders, the quantitative study discovers that improved TC, increased EI, and a supportive OC all have a favorable influence on ASP. SC also partially mediates the effects of TC, EI, and OC on ASP. The study offers significant insights for SMEs looking to include sustainability into technology initiatives by concentrating on these factors and actively incorporating stakeholders to develop a more environmentally conscious future.