Proceedings of Trends in Electronics and Health Informatics

Depression is a serious mental disorder that affects millions of people around the world. One of the challenges in dealing with depression is to detect it early and accurately. In this study, we use electroencephalography (EEG) to measure the brain activity of depressed and non-depressed individuals. We aim to find out which EEG features can serve as reliable biomarkers for depression recognition. We processed the raw EEG data from 128 electrodes by selecting 16 relevant electrodes, removing outliers and wrong formatted data. Then we applied a Gramian angular field (GAF) transformation to the relevant electrodes. We fed the resulting matrix into a recurrent neural network (RNN) model with two hidden layers using ReLU activation function and a sigmoid activation function as output layer, which produces binary results. We used this model to classify 53 patients with different brain disorders based on their EEG features (16 electrodes * 3 features). The comparison of the classification accuracy of our proposed work with other dataset is found to be better. It is observed that the overall depression classification performance is found to be very promising as compared to previous research works being conducted on both the datasets with an accuracy of 90.57%. In this paper, we have presented a novel approach to identify depression from EEG signals. We applied RNN to these features to classify depressed and non-depressed subjects. Our approach is based on empirical experiments to find the best parameters for our model. This may require some trial and error to achieve the optimal results, which could be seen as a potential drawback of our method. We plan to address this issue by using an optimization algorithm to automate the hyperparameter selection process in our future work.

In this digital era, almost everyone can share their thoughts over the internet through texts in messages, posts or blogs, and people often convey emotions of different sorts as well. Sometimes it is about mental health, review about a bad product or simply opinions on different aspects of the modern world. Because of its massive popularity, emotion detection from text has become a burning topic for researchers. This review paper tries to provide understanding of different levels in various emotion models, different works that have been done in emotion identification from text and how it works along with introduction to various available datasets. Also how different models like LSTM, CNN, BERT and so on are utilized by researchers to provide better results in text-based emotion detection is also discussed with their advantages and disadvantages. Finally this paper also points out different obstacles and challenges that lie in the study of emotion extraction from text.

The recent outbreak of monkeypox spread over seventy-six countries, seventy of them not having any prior history of monkeypox cases, because of its complex transmission patterns and high frequency of human-to-human transmission. As a result, it has appeared as one of the most critical health concerns. Therefore, a quick diagnosis is crucial, and in this case, computer-aided lesion detection may help identify suspected cases quickly. In this work, the first three pre-trained architectures, VGG19, ResNet50, and DenseNet121, for the transfer learning approach. Among them, ResNet50 proved a comparatively ideal outcome and diagnostic validity arrived at 97.68%. Then, the system combines the in-depth features and machine learning classifiers to get more effective results. From the experimental outcomes, the research finds that the detection accuracy of ResNet50 + SVM is 99.55%, which is improved by 2.47% from the baseline ResNet50 model. In addition, our proposed system also achieves 99.82% sensitivity, 99.33% specificity, and 99.69% AUC, respectively. Therefore, this research shows that the introduced method can be a beneficial tool for clinical decision-making.

Images are vague because they can be considered as sets of bright data, and brightness is a vague concept. Soft computing is a computational paradigm designed to handle and process ambiguous data. Fuzzy mathematics plays an important role in soft computing. Fuzzy techniques can effectively address the vagueness of an image. This paper proposes a nonlinear fuzzifier (NFr) and a time-efficient as well as effective fuzzy technique for image contrast enhancement with fine details preservation. First, the suggested technique fuzzifies an image with the proposed NFr. Then, it defuzzifies the fuzzified image with the inverse of a linear fuzzifier for producing the final image. The technique effectively enhances the contrast of an image and preserves its fine details by reducing its vagueness within two simple steps in a very short time. The technique has been compared with some conventional and cutting-edge contrast enhancement methods (both fuzzy and crisp) regarding eight objective image quality assessment (IQA) measures: linear index of fuzziness, discrete entropy, mean brightness preservation, bit-plane to bit-plane similarity, SSIM, PSNR, BRISQUE, and contrast improvement. Also, elapsed time and visual IQA are carried out for subjective IQA. The findings indicate the time efficiency and effectiveness of the proposed fuzzy technique in image contrast enhancement. The source code is available at https://doi.org/10.13140/RG.2.2.14716.51849 .

The Internet has transformed into a hub for a wide array of illegal activities, ranging from annoying spam ads to financial scams, all thanks to advancements in modern technology. With the constant enhancements in network technology, data traffic on the Internet is ballooning in both volume and scope. Consequently, the complexity and prevalence of cyber threats and attacks are on the rise, with websites becoming the prime targets for hackers. These intruders, or hackers, embed harmful content into web pages with the intent of carrying out nefarious activities. They employ various tactics, such as hacking into computers or trying to harvest information through compromised or malicious websites. Some of these tactics involve injecting malicious software into web links. Therefore, it has become more critical than ever to identify and block such websites before the average user can stumble upon them. In this paper, we propose a method that utilizes machine learning techniques to detect fraudulent websites. We evaluate the prediction accuracy of different machine learning classification methods in this study. We trained our model using various algorithms, including Naive Bayes, Support Vector Machines (SVM), K-Nearest Neighbors (K-NN), Decision Trees, Random Forests, and AdaBoost. Ultimately, we employed an ensemble approach and achieved an outstanding accuracy rate of 99.98%, surpassing other methods in effectiveness.

Language-based multilabel text classification is a crucial topic in natural language processing (NLP). However, unlike English, the development of natural language processing in Bengali is still in its infancy, and there is little research done specifically for the Bengali language, one of the most widely spoken languages in the world. As a consequence, it’s past time to address this issue for effective information management and data structure. This research presents a novel human-annotated Bangla sentence dataset. The dataset was categorized into five classes (assertive, interrogative, imperative, optative, or exclamatory) containing 10,000 sentences based on the function and purpose of Bangla grammar context. Fine-tuning applied six transformer-based pre-trained multilingual and monolingual BERT models to this multilabeled dataset. The study evaluated the effectiveness of various models, including BERT, AlBERT, RoBERTa, DistilBERT, XLNet, and BanglaBERT, in sentence classification tasks. The results showed that BanglaBERT outperformed all other models with an accuracy of 99.03%. The study concluded that BanglaBERT is superior in achieving exceptional accuracy and precision, surpassing the capabilities of other popular machine learning models such as LSTM, RNN, SVC, DT, KNN, and RF, as well as other BERT models of Bangla text classification. BERT-based models have great potential in understanding complex language nuances and context and are critical to advancing natural language processing tasks, especially in Bangla language processing.

The dynamic potential of Convolutional Neural Networks (CNN) enhanced by Transfer Learning offers a promising avenue for object recognition within the realm of agriculture. To explore this potential in the context of Bangladeshi agriculture, this study leverages CNN enhanced by Transfer Learning to identify Bangladeshi-grown fruits from a dataset featuring 24 distinct types, including Mango, Jackfruits, and Lychee, among others. Utilizing an extensive dataset that encapsulated high-quality images of these fruits, we embarked on a comparative analysis of six advanced CNN architectures: VGG19, Xception, ResNet152V2, MobileNet, DenseNet201, and NASNetLarge. The results were illuminating. While VGG19 lagged behind with an accuracy of 48%, most models showcased impressive performance, with DenseNet201 emerging as the frontrunner at a remarkable 95% accuracy. The findings not only underscore the transformative role of Transfer Learning in optimizing CNN performance, especially when faced with limited labeled datasets, but also spotlight the applicability of these models in revolutionizing quality control and management within the Bangladeshi agricultural sector. As the demand for automated and accurate agricultural systems gains traction globally, the insights from this research offer a foundation for further exploration and application in similar contexts, demonstrating the scalability and adaptability of Transfer Learning-enhanced CNNs.

Fake news detection is a critical challenge in the digital age, where misinformation spreads rapidly, causing real-world harm. In the context of the Bangla language, this problem is exacerbated by the scarcity of labeled data for model training. This paper introduces an enriched dataset of Bangla fake news, containing 7000 authentic and 1000 fake news texts, meticulously labeled for comprehensive analysis. To tackle this issue, we explore state-of-the-art language models such as DistilBERT and RoBERTa. Our experiments with these models achieve remarkable accuracies of 94.5% and 94.3%, respectively, demonstrating the potential of transformer-based architectures for fake news detection in Bangla. Additionally, we propose a novel Deep Convolutional Neural Network, named “BFakeNewsCNN,” which, despite achieving a respectable accuracy of 85.7%, offers an efficient alternative for low-resource settings. This research contributes to the advancement of Bangla fake news detection, addressing the challenges of limited data resources while paving the way for more robust and accurate detection systems.

Performance evaluations are conducted with the intention of establishing each worker’s level of commitment to the company. Numerous businesses are challenged by the issue of employee attrition, which occurs when talented workers with years of expertise leave the firm on a regular basis. On the other hand, employee turnover can be caused by a wide variety of factors, and it can be challenging for the HR manager or the head of each department to recognize warning indications in a timely manner. Forecasting the performance of employees is a vital part of a successful business. As a result, we provide a methodology for anticipating employee attrition in this study so that we can carry out talent management strategies that were previously done ex post. A model of predication for employee attrition was built for this study using 30 factors. We use 1470 entries from the “IBM HR Analytics Employee Attrition and Performance data” for this purpose. We utilize the two-stage staking ensemble model, which integrates the basic models of Random Forest, K-nearest Neighbor, Naive Bayes, and Decision Tree with the meta-model of Logistic Regression for forecasting. Our suggested two-stage stacking model has an accuracy of 88.01%. Additionally, we compared our suggested model to Decision Trees, Support Vector Machines, and Gaussian Naive Bayes, three major machine learning model.

Alzheimer’s disease is a significant public health concern, and early detection is crucial for effective intervention. In this paper, we explore the application of ensemble learning approaches to classify Alzheimer’s disease in brain imaging data (MRI images). We employed several pre-trained deep learning models, including VGG-19, ResNet-152, EfficientNetB1, and EfficientNetB2, to extract valuable features from the imaging data. These models were individually trained for ten epochs, resulting in impressive training and validation accuracies. Specifically, VGG-19 achieved 99.22 and 93.88% accuracy, ResNet-152 achieved 98.64 and 92.71% accuracy, EfficientNetB1 achieved 90.04 and 86.57% accuracy, and EfficientNetB2 achieved 93.28 and 93.29% accuracy. To further improve classification performance, we constructed two ensembles, denoted as Ensemble-1 (VGG-19, ResNet-152, and EfficientNetB1) and Ensemble-2 (VGG-19, ResNet-152, and EfficientNetB2), by combining the individual models. These ensembles exhibited remarkable accuracy, with Ensemble-1 achieving 99.99% training and 95.04% validation accuracy, and Ensemble-2 achieving 99.81% training and 97.16% validation accuracy. The results highlight the potential of ensemble learning in enhancing the accuracy and robustness of Alzheimer’s disease classification, offering a promising avenue for early diagnosis and intervention.

Data sorting is a primary problem of computational sciences. There exist different sorting techniques among them bubble sort is a primitive one and it is widely used in many applications due to its simple architecture and development procedure. However, the main concern of this algorithm is time consumption. So, different researchers aimed to reduce its complexity, and modified versions are available such as insertion, and cocktail shaker. Still, challenges exist and therefore in this research, a novel method is proposed for data sorting based on cocktail shaker sort where unnecessary swaps are reduced by checking similar values in neighbors and skipping iteration for these values like chess board knighted approaches. Hence, it reduces the number of swaps as well as time. The method is tested in three datasets and achieves a superior performance in terms of the number of swaps as well as time. It shows 81–90% improvement in the case of swaps and 40–70% superior performance in the case of CPU time. Hence, the algorithm can be used for iterative sorting. Code and datasets are available at https://github.com/niloy-999/Modified_sorting_dataset .

There are now many more people sharing their views, ideas, and opinions because of the proliferation of user-generated material on the Internet and the development of information technology. People frequently use social media platforms like Facebook, Instagram, WhatsApp, and Twitter to express their thoughts and feelings. Opinion analysis, which is the act of identifying emotional undertones from this enormous volume of Internet data, has become more important as a way of understanding public opinion, trends, and brand impressions. However, because of the distinctive features of the network, such as informal language, character constraints, and the quick influx of data, conducting sentiment analysis on Twitter data poses unique obstacles. By considering the above difficulties, we have proposed a machine learning-based approach to classify the emotions as either positive, negative, or neutral. This paper conducts a thorough analysis of sentiment on Twitter data using the Support Vector Machine (SVM), Maximum Entropy (Max Ent), and Naive Bayes Multinomial (NBM) machine learning algorithms. To do this experiment, we deal with the large volume of data collected from Kaggle. To train the machine learning model, this wide range of tweet datasets from different subjects and domains is preprocessed using the Natural Language Tool Kit (NLTK). The results from these models were tested using various testing matrices like precision, recall, and F1-score. We achieved a maximum accuracy of 97% for SVM compared to all three classifiers.

Road object detection is a critical aspect of developing assistive technologies to enhance the mobility and safety of visually impaired individuals. In the context of Bangladesh, a densely populated and diverse environment, the need for accurate and contextually relevant road object detection is particularly significant. This paper introduces a novel road object detection dataset tailored specifically for the Bangladeshi context, aimed at facilitating the advancement of computer vision systems for assisting visually impaired pedestrians. The proposed dataset, named “BanROD” (Bangladesh Road Object Detection Dataset), is constructed by capturing images across Sylhet, Bangladesh. We primarily constructed this dataset which is trained on a baseline machine learning model. The dataset is annotated with meticulously labeled bounding boxes around road objects of interest including persons, CNGs, cars, upstairs-downstairs, and other obstacles. To ensure diversity, images are collected under different lighting conditions, weather scenarios, and traffic densities. In addition to dataset creation, we train and evaluate a deep learning-based object detection model “MobileNetV2” on the BanROD dataset. We then evaluate the model’s performance by measuring loss and average precision to find out how well it performs on real-world images. This solution is only for mobile devices as they are more handy for blind people around us.

Operating systems, which make computing easier, are a crucial part of computing. To make this possible, a plethora of operating systems exist, each of which differs from the others. Some provide more facilities than others, while some are developer-friendly but not beginner-friendly. Some systems even lack common features, which is unfortunate. It would be better if all features could be merged into a single system in an intuitive way. The user system would thus be a feature-rich and beginner-friendly operating system, as well as a security- and performance-optimized system. This is why my own implemented and customized Linux system, NewBreeze, is presented here to fulfill user requirements. Since the source code of Linux systems is open, it can be customized as needed. This is also the reason for choosing the Linux system as the basis for my system. In order to fulfill the requirements of the developed system as mentioned in the contribution, I have also slightly customized the Linux Zen kernel and the base Arch Linux system as needed. To create the user ISO file, the “archiso” package and KDE’s drop-down terminal “Yakuake” were utilized. A brief and complete implementation procedure has also been provided in this article. Shell scripting was utilized mostly in the source code. JSON format is used for a lot of configuration tasks. Other languages utilized are Python, C, C++, and Assembly. The “waydroid” and “waydroid-image-gapps” packages were used to implement the Android subsystem, and the “wine”, “wine-mono”, “wine-gecko”, and “winetricks-git” packages were used to implement the installation process of Windows software. All of the aforementioned packages were obtained from the official Arch Linux repositories, as well as from the “aur” and “chaotic-aur” repositories. Finally, the command “mkarchiso -v.” was executed to build the ISO file.

In this research initiative, our central objective is the development of an emotion recognition system through the utilization of cutting-edge deep learning methodologies. Our methodology involves adopting a subject-independent approach to classify emotions based on Electroencephalogram (EEG) signals sourced from a standard benchmark DEAP dataset. To ensure data quality and reliability, we initiate our process by meticulously preprocessing the raw signal data, which includes the application of Normalization and Common Average Reference (CAR) techniques. Subsequently, we employ Discrete Wavelet Transform (DWT) technique to extract salient features from the cleaned EEG data. These extracted features serve as the foundation for training three distinct deep learning models: the CNN-LSTM, CNN-GRU, and 2D-CNN models. To consolidate their predictive capabilities, we employ a Majority voting algorithm, effectively combining the strengths of these models. Notably, our proposed deep ensemble learning approach yields an impressive accuracy rate of 88% when evaluated on the challenging DEAP dataset.

Pests and insects pose a significant threat to global agriculture, causing crop damage and quality loss, with a massive annual costs. Early detection of these agricultural challenges is essential for sustainable farming and food security. Challenges in early pest detection include class imbalance, computational limitations, and the need for comprehensive feature learning. In response, our study introduces “Tri Focus Net (TFN),” a hybrid model that seamlessly integrates multiple attention mechanisms, including Channel, Soft, and Squeeze-and-Excitation Attentions, effectively addressing these issues. This combination empowers TFN to discern critical details both globally and locally within images, significantly enhancing feature extraction and robustness. Additionally, we have successfully integrated DenseNet 201 with TFN, further enhancing our model’s feature extraction capabilities and robustness. In our work, we achieved exceptional results, attaining a maximum accuracy of 94.20% for the insects dataset and a perfect 100% accuracy for the pest dataset. Our contribution to agriculture lies in providing a cost-effective, precise, and robust solution for pest and insect detection, underpinning the sustainability of crop health and the agricultural industry.

Email is one of the most essential and useful communication channels of this advanced world. The continuous growth of email user has led to a massive rise of unsolicited emails also known as spam emails. Hence, appropriately managing and classifying this huge number of emails pose a critical challenge. Despite frequent attempt in creating more reliable and effective solution, there is room for improvement. In this paper, an efficient spam email detection technique is proposed and examined on a combined training dataset after applying feature engineering techniques. Thereafter the proposed model is tested using combination of five different datasets and several machine learning algorithms including SVM, SVC, Naïve Bayes, KNN, logistic regression, decision tree, etc. Experimental result shows high detection rate while reducing training time. Comparative analysis highlights the remarkable outcome of this experiment as compared to existing literature. The overall accuracy achieved by the suggested approach is 99.39%.

This paper addresses the severe threat of White Spot Disease (WSD) to the shrimp farming industry in Bangladesh and proposes a comprehensive methodology for predictive modeling and early detection using advanced machine learning techniques. The study emphasizes the importance of shrimp aquaculture in Bangladesh, outlines the impact of WSD outbreaks, and sets objectives for developing a specific prediction model for shrimp farming. Employing Random Forest, Multinomial Naive Bayes, and Bagging with Decision Trees, the research achieves impressive accuracy rates exceeding 90%, with the Bagging with Decision Trees classifier and Random Forest reaching a maximum accuracy of 97.87%. The outcomes hold significant implications for early WSD identification in the shrimp farming industry, contributing to Bangladesh’s GDP and providing valuable insights for managing aquatic. The disease outbreaks globally.

Vehicle license plates (LPs) detection and recognition is a fundamental technique in computer vision that plays a crucial role to extract, localize, and authenticate the information of a vehicle. This paper presents an efficient method for Bangla LPs recognition and vehicle authentication using edge analysis and pattern matching algorithm in real-time nature. This method performs four modules as: preprocessing, license plate identification, character recognition, vehicle authentication. A multi-step preprocessing technique has been applied to extract and understand the potential information in case of any uneven images. To identify edge connectivity and uniformity, Canny edge detector provides more promising in contour analysis and character localization. The segmented characters are converted into texts and verified using the pattern matching algorithm by a predefined database. For this research, a database consisting of 3000 automobile license plate photos captured under different climatic circumstances has been compiled. Additionally, we have curated two additional databases consisting of templates for recognition and registered automobiles for the purpose of authentication. The proposed model can achieve accuracy of 96% in LPs detection, 99% in character recognition, and 100% in vehicle authentication by extracted texts from LPs.

The most occurring cancer worldwide is breast cancer in which abnormal breast cells grow and form tumors. If left unchecked, it can be more fatal and can spread throughout the body. Globally, 2.3 million cases have been diagnosed and there were 685,000 fatalities in 2020, according to the World Health Organization (WHO). To overcome the fatalities, an early diagnosis is the best approach. Manual diagnosis is not an easy process for this cancer, using mammography images, and always requires an expert person. In this paper, we have used a VGG16-based model to eradicate the feature from the intermediate layers and concatenate the intermediate layers. Then, we used a Convolutional Neural Network (CNN) to train the feature and predict the accuracy and classification. We have evaluated our performance analysis with confusion matrix analysis, ROC curve analysis, and other methods. In our results, we have accomplished an accuracy of 94% among the datasets. Comparing our suggested framework to state-of-the-art (SOTA) technology demonstrates that it increased accuracy, and additional study may enable it to be enhanced to decrease false positives and false negatives in screening mammography findings.

Antibiotic resistance is a significant challenge in medicine, stemming from the misuse of antibiotics and resulting in drug-resistant bacteria. To address this, it is crucial to identify appropriate treatment options, including Over-the-Counter (OTC) medications and doctor consultations, to minimize unnecessary healthcare costs while effectively managing antibiotic resistance. Developing a machine learning-based prediction model for OTC medications and doctor consultations becomes imperative. The research utilizes patient demographic data, medical histories, and symptom data labeled with OTC drug usage or doctor consultations from CMED, a large healthcare system in Bangladesh. An ensemble model has been created to predict treatment options, achieving an accuracy of 94%, precision of 91%, and an impressive ROC value of 0.98 in determining whether patients should opt for OTC drugs or seek doctor consultations. These findings highlight the value of machine learning in optimizing patient care and healthcare outcomes, guiding patients to appropriate care and reducing the excessive use of antibiotics. This approach holds the potential to significantly impact the development of antibiotic-resistant bacteria and improve patient outcomes.

Cardiovascular disease is currently a prominent sickness that kills the majority of sufferers. The medical evaluation of cardiac disease presents significant problems. This diagnostic method is complex, requiring precision and efficiency. Early detection of heart disease can significantly lower the chance of death. In light of the high occurrence of cardiac issues in contemporary times, the prediction of heart disease has emerged as one of the most challenging endeavors within the medical field in recent years. Scientists examined a plethora of closely related characteristics to identify the most trustworthy predictors of these diseases. In this work, we use Machine Learning (ML) approaches to detect the existence of cardiac problems. The suggested method predicts the likelihood of heart disease and classifies people into risk categories. This is performed by utilizing several machine learning algorithms such as Support Vector Machine (SVM), Gradient Boosting, Random Forest (RF), K-Nearest Neighbors (KNN), Naive Bayes (NB), and Logistic Regression (LR). The developed system is trained and evaluated using a composite dataset derived from two separate sources. The results of the experiments show that, when compared to other machine learning algorithms, the Random Forest approach achieves the greatest accuracy rate, achieving an amazing 99.99%.

Cervical cancer has been one of the leading causes of female early death in recent years. In developing countries, cervical cancer occurs in more than 85% of cases. Cervical cancer has been associated with several risk factors. For forecasting the prognosis of cervical cancer patients, we created a prediction model in this study based on early screening and risk trends in individual health records. In this paper, we analyze the risk variables for cervical cancer using ML classification methods. The study’s dataset is very unbalanced and includes missing values. Consequently, the ROS approach, a sampling method was used. The effectiveness of class imbalance was demonstrated by comparing the suggested model’s accuracy, sensitivity, and specificity. The RF, GB, and MLP machine learning models perform better with 99.07, 98.57, and 98.51% accuracy when using the Random Oversampling technique, and XGBoost.

The convergence of machine learning and medical data presents an exciting frontier in the realm of healthcare, with the potential to revolutionize the early detection of diseases. In this study, we introduce innovative machine learning models designed for the early prediction of three critical ailments: diabetes, heart disease, and liver disorders. To enhance the performance of these models, we rigorously fine-tuned their hyperparameters, a critical aspect of the model development process. Our approach involved the utilization of various classification algorithms, such as logistic regression (LR), extra tree (ET), support vector machine (SVM), Naïve Bayes (NB), decision tree (DT), and random forest (RF). Furthermore, we employed ensemble learning techniques like bagging and boosting, using the aforementioned traditional algorithms as base estimators. All these algorithms underwent extensive hyperparameter tuning to optimize their predictive capabilities. To assess the performance of these models, we conducted a thorough tenfold cross-validation, enabling us to make a comprehensive comparative analysis and identify the most effective models for each dataset. Notably, our efforts bore fruit with exceptional results. For instance, we achieved an impressive accuracy rate of 99.22% in predicting diabetes using the traditional SVM classifier. In the case of the Statlog heart dataset, we reached an accuracy of 85.67% by utilizing the random forest classifier within a bagging ensemble. In predicting liver disorders, we achieved a 73.75% accuracy by employing both boosting random forest and boosting extra tree classifiers. Additionally, we elucidated the reasons behind the variation in results, providing valuable insights. These experimental findings underscore the superiority of our proposed models over existing methods in terms of predictive accuracy. Consequently, our research represents a significant step forward in the early diagnosis and prevention of diseases within the realm of healthcare.

The monkeypox virus can spread to humans but is not as dangerous as the smallpox virus. Monkeypox is usually found in the wild forests of Africa. Since the COVID-19 pandemic started, it has spread to more places around the world. Animals, like different kinds of rodents and nonhuman mammals, are the main hosts. People are now afraid of it because of how quickly it is spreading. They think it could be the next COVID-19. So, it is important to identify the virus at the initial stage before it starts spreading globally. In this work, we have developed several CNN models to identify the Monkeypox virus from other skin diseases. For this purpose, 6 CNN models were trained and compared, consisting of mainly VGG, DenseNet 121 and 201, Inception, Xception and MobileNet. Tensorflow was used along with the CNN models for better classification of the images. For this classification, two different datasets were used for training and testing, were a set of random images were used for the testing purpose. Inception V3 classified the disease with the precision of 94.99% and 89.97 as the F1-score.

Skin cancer poses a significant threat to human health, particularly when early identification and accurate diagnosis are lacking. Early detection is pivotal for successful treatment, given the potential lethality of this disease. While skin cancer is commonly associated with sun-exposed areas, it can also manifest on regions shielded from sunlight. In 2020, there were 150,000 new melanoma cases reported worldwide, making it the 17th most prevalent form of cancer. With a rapid increase in skin cancer cases, there is an urgent need for early and precise differentiation between cancerous and non-cancerous lesions. In response to this critical demand, we propose an advanced model for region-based skin cancer lesion classification utilizing the Faster CNN architecture. This innovative model aims to revolutionize skin cancer diagnosis, reducing reliance on dermatologists’ experience and conventional diagnostic tools. Our study presents an efficient automated system with enhanced evaluation and accuracy metrics, surpassing both prior research and expert dermatologists, achieving an impressive 92% accuracy for the proposed region-based Faster CNN model. This research opens the door to a more effective approach to skin cancer classification and early diagnosis.

Brain tumors, particularly those of the highest grade, can substantially lower life expectancy. A precise diagnosis is required for individuals with brain tumors to receive the best possible care. The conventional way of diagnosing brain tumors is Magnetic Resonance Imaging (MRI) while recent tumor detection models leverage Convolution Neural Networks (CNN) and pre-trained models, many prioritize accuracy over providing explanation and interpretability. This paper addresses these shortcomings by proposing an innovative approach to the categorization of brain tumors by exploring transfer learning and Explainable Artificial Intelligence (XAI). The study uses the top five pre-trained models to classify three different types of tumors: VGG16, ResNet50, Xception, InceptionV3 and DenseNet201. Notably, a customized VGG16 model obtains a promising test accuracy of 99.04%, which is comparable to top literature scores. A significant contribution of this work is the integration of the top three explainable AI tools: Gradient-weighted Class Activation Mapping (Grad-CAM), Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP). These tools enhance visualization and provide valuable explanations for model predictions.

While the classification of medical data poses a formidable challenge, it remains a source of profound fascination within the research community. Its potential to enable precise preventive measures against future ailments is a driving force. Machine learning (ML) algorithms play a pivotal role in this endeavor, and the effectiveness of these algorithms is deeply intertwined with the pre-processing of data. This study seeks to evaluate a range of ML-based models with a specific focus on their utility in predicting Autism Spectrum Disorder in Toddlers (ASD-T) and Chronic Kidney Disease (CKD). The pre-processing techniques encompass a spectrum of procedures such as imputing missing values, feature selection, resampling, and feature scaling. In the pursuit of precision, hyperparameter tuning is employed to fine-tune these models. Subsequently, six classification techniques—Logistic Regression (LR), Support Vector Machine (SVM), Naive Bayes (NB), Decision Tree (DT), Extra Tree (ET), and Random Forest (RF)—are executed using a rigorous tenfold cross-validation approach to classify the datasets. The integration of ensemble learning techniques, namely bagging and boosting, enhances performance while safeguarding against bias and overfitting. Although the literature is replete with experimental examples, room for improvement persists. This study distinguishes itself by achieving significantly superior results compared to prior research. Notably, our proposed model, the traditional ET classifier, delivers unparalleled performance, boasting a remarkable 100% accuracy in the ASD-T dataset and an impressive 99.85% accuracy in the CKD dataset.

Lung cancer is a specific kind of malignancy that arises in the tissues of the lungs, specifically in the cells that comprise the lining of the respiratory passages. Smoking is well recognized as the primary etiological factor contributing to the development of lung cancer. The inhalation of secondhand smoke, as well as exposure to asbestos or radon inside one’s own residence, Being employed, and having a familial predisposition to the ailment are further factors. Factors contributing to the chance of developing lung cancer. In their lifetime, 1 in 16 men and 1 in 17 women were diagnosed with lung cancer. The variability in survival rates may be attributed to the extent of illness dissemination at the time of its identification. Early detection may have substantial benefits. The term “impact” refers to the effect or influence that something has on a particular situation. The dataset was collected from a website. The study used a global-based Particle swarm Optimization (PSO) algorithm for feature selection. With our approach, we want to improve the dataset’s statistical feature accuracy and predict lung cancer accurately based on the given data. Among the employed classifiers, the Support vector machine and Naive Bayes obtained maximum accuracy of 97% with 8 out of 16 features.

Cardiovascular diseases (CVDs) are considered as significant global health issue as they bear the brunt of the world’s morbidity and mortality. This research paper aims to provide a comprehensive analysis of cardiovascular illnesses’ causes, warning indications, and available treatments. The multifactorial nature of CVDs, which encompass various diseases for example heart failure, stroke, coronary artery diseases, and peripheral vascular disorders, is explained in the first section of the study. A thorough analysis of the pathophysiological mechanisms behind diverse disorders sheds light on the complex interplay of genetic, behavioral, and environmental factors impacting the emergence of certain disorders. In this study, we have assessed and contrasted various machine learning-based techniques including random forest (RF), logistic regression (LR), support vector machine (SVM), and neural networks in order to predict cardiovascular diseases. Researchers and healthcare professionals can gain valuable insights into the applicability and efficacy of each strategy by carefully examining the advantages, limitations, and associated performance indicators of each method.

Hepatitis C is a global health concern that can spread through blood from infected people directly. The patient can avoid the danger of mortality by detecting this disease early. In this study, we use laboratory values and demographic information to predict Hepatitis C using 10 machine-learning algorithms such as Random Forests, Logistic Regression, Decision Trees, Support Vector Machines, Naive Bayes, K Nearest Neighbors, Gradian Boost, Boost, Artificial Neural Network and Cat Boost and 7 ensemble algorithms are GB-XGB, LR-XGB, LR-SVC, LR-KNN, LR-GB, LR-RF, SVC-XGB. Machine Learning algorithms are. This study aims to contribute to the medical sector. We get an accuracy of 99% using Logistic Regression, ANN, and SVC algorithms. Similarly, we achieve 98% accuracy using ensemble algorithms. This result demonstrates the effectiveness of our approach.

Engagement assessment plays a crucial role in understanding and improving the educational experiences of students with autism spectrum disorder (ASD). Traditional methods of engagement assessment heavily rely on subjective observations and manual coding, which can be time-consuming, prone to bias, and lack of precision. In this study, we propose a deep learning-based approach, specifically convolutional neural networks (CNNs) to automatically analyze multimodal data streams, including facial expressions and body movements collected during classroom activities. The multimodal data is preprocessed and fed into CNN for feature extraction, followed by CNN-2 for temporal modeling and engagement prediction. The results demonstrate that the proposed deep learning-based approach achieves significantly higher accuracy that is 97.1%. In the future, an Intelligent Tutor System (ITS) will be built to help children with ASD study more effectively via the use of mobile or online applications. Evaluating participation from since engagement is a cognitive trait shared by humans, visual signals alone are insufficient. Physiological elements must thus be present. In the future, we will work on a model that can evaluate a student’s degree of involvement by combining behavioral and physiological data.

Cotton trees are tropical and subtropical trees that flourish in warm climates. It is a popular, costly as well as a cash crop. Farmers face difficulties selling their products when their production is decreased due to diseases harming cotton plants. It is critical to manage any dangerous illnesses as soon as feasible in order to increase quality and production. This problem prompted the development of novel technologies for detecting and diagnosing cotton plant diseases, as well as expert systems for disease prevention. The use of computer vision in agriculture has grown in popularity as a result of its ability to give critical information in real time. To achieve our goal, we built a dataset that included 4 classes of diseased and leaf images. We have applied InceptionV3, DenseNet, and MobileNetV2 to our dataset. By using DenseNet, we achieve the best accuracy of 99.24%. Our system with the capability of identifying leaf diseases of cotton has been built. As a result, we believe that our initiative will help the user by allowing us to make more items, which will impact our economy.

The study uses powerful learning methods to identify powdery mildew and leaf scorch on strawberry fruit and leaves. The suggested solution combines ResNet50 and InceptionV3 cognitive neural networks. This strategy involves collecting powdery mildew and leaf scorch datasets and improving them. ResNet50 and InceptionV3 extract discriminative features and detect powdery mildew and leaf scorch patterns. The aggregate characteristics are put into a fully connected layer and a softmax classifier for illness classification. An optimized technique adjusts the model’s hyperparameters during training for optimal classification performance. Comprehensive testing shows that the combined model beats ResNet50, InceptionV3, and other illness diagnostic approaches. Integration of information utilizing fusion approaches improves representation and illness diagnosis. Research suggests that the reinforcement learning-based approach can identify powdery mildew and leaf scorch in strawberry fruits with an accuracy rate of 98.42% and a validation rate of 99%. This technology impacts strawberry crop disease control.

Bangladesh is one of the countries that have various kinds of diseases among its vast population, throughout the year in different places. It is becoming a very crucial factor in the medical sector to be cautious regarding the health of people. In this paper, a prediction model is provided to represent the geo-temporal disease prediction model. This model will help people to be aware of their health according to location and season. We made a visualization representation that can demonstrate the disease prediction model. The proposed disease prediction model is constructed in such a way that it can find out the occurrences of the inputted disease in a given time and place. By implementing the algorithm on the mentioned medical data, missing values were replaced by our proposed missing value filter algorithm SVM. Then from the historical medical data, we analyzed the frequently occurring diseases along with geo-temporal relations. Moreover, after training the system with our specified analysis, the future disease prediction model was made with respect to the geo-temporal relation. We revealed the empirical result by which our application is useful and can be used in the real world for predicting future diseases in Bangladesh using geo-temporal location.

One of the most delicate concerns is cyberbullying due to today’s worldwide web advancement. Bullying may contribute to severe consequences, including mental health issues, academic performance challenges, and job dropout. It can also instill hatred in individuals or communities, making cyberbullying prevention critical. Our main goal in this paper is to mitigate the impact of cyberbullying and contribute to fostering a healthier Internet environment where individuals can interact without fear of unwarranted criticism. We propose a proactive approach involving early detection mechanisms. In this study, we employ a comprehensive analysis of tweets using multiple factors. The data is preprocessed using machine learning methods and natural language processing (NLP), followed by Term Frequency Inverse Document Frequency (TF-IDF) and Count Vectorizer. Several machine learning models, including Light Gradient Boosting Machine (LightGBM) and 12 others, are evaluated. The dataset comprises over 47,000 tweets from various perspectives. Our analysis indicates that the LightGBM classifier outperforms other models with an accuracy of 94.24%, precision of 95%, recall of 94%, and an F1-score of 94%.

Jackfruit is the national fruit of Bangladesh, and one of the most consumed fruits in India, Sri Lanka, Philippines, Indonesia, Malaysia, Australia, and many more countries. The every year due to diseases jackfruit production stays lower than the expectation. The early identification of these diseases is crucial for agricultural productivity, and our research showcases the effectiveness of deep learning techniques in addressing this issue. This study introduces an innovative approach to detect and classify two common jackfruit leaf diseases, algal spot and black spot, combining the YOLOv8 object detection model and the EfficientNetB7 pre-trained deep convolutional neural network model. The models trained on 6,360 images of those two diseases and our methodology achieved an accuracy of 99.9% on training and 99.5% on testing. This work represents a valuable contribution to agriculture, emphasizing the synergy between cutting-edge technology and age-old challenges, with the potential to positively impact the agricultural industry and global food security.

Beauty products are an integral part of daily routines for many women. The beauty industry continually evolves, introducing innovative formulations and trends that cater to diverse preferences and needs. Sentiment analysis of beauty products is crucial as it allows companies to understand customer opinions, preferences, and experiences. In this study, we explore sentiment analysis in the context of beauty products. We collected data from various platforms like shajgoj.com, kireibd.com, and klassy.com.bd, resulting in a dataset of 4431 reviews. Employing this unique dataset, we conducted a comprehensive analysis, employing a trio of neural network architectures: the foundational simple neural network, the robust 1D convolutional neural network (1D CNN), and the intricate long short-term memory (LSTM). Upon completing data preprocessing, we incorporated word embeddings derived from the GloVe 100D file, resulting in the attainment of an accuracy level of 78.78%. By combining data collection, model experimentation, and embedding techniques, we gain a multifaceted view, offering deeper insights into beauty product sentiment. In addition to advancing sentiment analysis methodologies, our study signifies a positive step toward empowering beauty industry stakeholders with actionable intelligence.

Agriculture, as an indispensable facet of human civilization and a vital economic contributor, plays a pivotal role in sustaining societies. Bangladesh, a nation heavily reliant on agriculture, currently employs traditional methods, resulting in a significant impact on its economic stability. Improvements in agricultural technology and practices have the potential to increase productivity, reduce waste, and improve the quality of agricultural products. As the global population swells and food demand rises, sustainable agricultural methods become increasingly crucial. To address this challenge, we have developed an AI-based precision farming system employing machine learning algorithm, including CNN networks, to analyze and predict crop health and potential yield. Additionally, we propose an AI-based decision-making module that forecasts crop yields, pest detection, and disease detection. Our system, rigorously evaluated using real-world data, has demonstrated substantial improvements in crop yield, fostering sustainability and contributing to global food security while minimizing environmental impact.

COVID-19 is a unique and devastating respiratory disease outbreak that has affected global populations as the disease spreads rapidly. Many deep learning breakthroughs may improve COVID-19 prediction and forecasting as a tool for precise and fast detection. In this study, the dataset used contained 8055 CT image samples, 5427 of which were COVID cases and 2628 non-COVID. Again, 9544 X-ray samples included 4044 COVID patients and 5500 non-COVID cases. MobileNetV3, DenseNet201, and GoogleNet InceptionV1 show the highest accuracy of 97.872%, 97.567%, and 97.643%, respectively. The high accuracy indicates that these models can make many accurate predictions, as well as others, are also high for MobileNetV3 and DenseNet201. An extensive evaluation using accuracy, precision, and recall allows a comprehensive comparison to improve predictive models by combining loss optimization with scalable batch normalization. This research shows that these tactics improve model performance and resilience for advancing COVID-19 prediction and detection and show how deep learning can improve disease handling. The methods suggested in this research would strengthen healthcare systems, policymakers, and researchers to make educated decisions to reduce COVID-19 and other contagious diseases.

Dengue fever remains a significant public health concern in Bangladesh, with recurring outbreaks posing substantial challenges to healthcare systems and communities. This study provides a concise overview of a comprehensive study aimed at unraveling the dynamics of dengue in Bangladesh through a synergistic combination of statistical and machine learning analyses. By applying statistical techniques, we first identify temporal and spatial patterns, uncovering seasonal trends, hotspot regions, and fluctuations in dengue incidence. The trend of safe childbirth practices gradually increased between 2000 and 2023. Dhaka, the capital city of Bangladesh, and its surrounding areas in the Dhaka Division showed a high number of dengue cases and deaths. The knowledge and awareness level about dengue was significantly higher for educated respondents (OR = 1.89, 1.21–1.97), residing in semi-urban regions (OR = 1.35, 0.93–1.41), female (OR = 1.39, 1.14–1.62), living in Dhaka division (OR = 3.72, 2.89–3.88), and housewife (OR = 1.52, 1.26–1.89). This initial analysis allows us to pinpoint high-risk areas and periods, facilitating targeted intervention strategies. In tandem with traditional statistical methods, we harness the power of machine learning to develop a predictive model which is capable of forecasting dengue outbreaks with enhanced accuracy. In conclusion, this study represents a comprehensive effort to deepen our understanding of dengue dynamics in Bangladesh. By combining statistical analyses with machine learning technique, we aim to provide actionable insights that can inform public health policies and interventions. Our findings have the potential to guide the allocation of resources, improve preparedness, and ultimately mitigate the impact of dengue fever in Bangladesh, offering a valuable framework for addressing similar challenges in other regions grappling with vector-borne diseases.

The official collection of numbers and letters shown on the front and rear of a road vehicle is called a license plate. License plate recognition is crucial for many applications, including traffic monitoring, vehicle tracking, and law enforcement. Automatic License Plate Recognition (ALPR) is one of the applications that benefited tremendously from convolutional neural network (CNN) processing, which is now the de facto processing method for complex data. This paper offers a comprehensive analysis and comparison of deep learning-based approaches for identifying license plates. The study also looks a dataset from Kaggle, pre-process the dataset for removing noise and artifact, utilized to train these models into test, train, and validation dataset. Additionally, the study examines the difficulties and prospective possibilities in deep learning-based license plate recognition, highlighting the potential uses of cutting-edge innovations such object detection using attention mechanisms and graph convolutional networks. In this study we utilized six transfer learning model which is VGG16, VGG19, DenseNet121, AlexNet, EfficientNet, and YOLOv4 where DenseNet121 provide the best or highest accuracy 98.66%. The goal of the study is to guide researchers and industry professionals toward powerful license plate detecting systems by employing deep learning approaches by way of this evaluation and analysis.

Bengali is the State Language of Bangladesh and about 300 million people use this language around the world. The maximum number of Bengali speakers are from Bangladesh. Though there is a common or Colloquial Language to communicate with each other in this country. But there are many Regional Languages in this country. Among the regional speakers one may not or do not understand other regional languages, that’s why it’s a crucial problem to communicate one regional speaker with another regional speaker. In this research, we are trying to make a platform for detecting one’s language region from his given sentences/words and according to his/her region, we are translating his/her sentences/words to Colloquial Language and this platform will train up him/his for learning the Standard Bengali Language as well as.

Human activities and traditional network infrastructure can be severely disrupted when natural disasters strike, making it difficult to maintain continuous connectivity. While natural calamities are beyond human control, their aftermath can be mitigated. In this context, ad hoc networks offer a promising solution as they do not rely on predetermined end-to-end connectivity. Following a natural disaster, providing emergency medical care to victims becomes imperative and can help save lives. This study introduces a post-disaster emergency health monitoring system capable of monitoring multiple health parameters such as ECG, body temperature, blood pulse rate, and blood oxygen levels. The system uses an ad hoc network to store and forward this data to a server. In cases where Internet connectivity is available, the data can be transmitted to the server using the Internet of Things (IoT) concept.

Nowadays, vast images are generated daily as we capture, transfer, and receive them in various sectors. Images have become a pivotal component of data in many different industries, contributing to decision-making, documentation, and artistic expression. However, verifying image authenticity has become more complex with the widespread availability of sophisticated software and tools that enable image alteration. As a result, determining whether an image is original or manipulated has become a complex task. In this paper, we propose an enhanced Transformer architecture to classify between original and manipulated images by using their metadata and EXIF data. Two datasets are built to train the framework. Each dataset carries metadata and EXIF data of original and manipulated images, respectively. An augmentation technique has been applied to ensure dataset balance and robustness. The proposed framework uses a parallel multi-head attention mechanism, which speeds up convergence throughout the training process and results in more efficient model learning. This versatile proposed framework can perform on different image formats such as JPG/JPEG, PNG, and BMP, highlighting its adaptability and real-world applicability. This framework has achieved 96.42% accuracy, showing its potentiality and capability to distinguish between original and manipulated images in this digital age.

In rural regions of Bangladesh, the scarcity of healthcare resources poses a substantial obstacle, frequently resulting in delayed diagnoses and inadequate medical care. This paper introduces an inventive solution, the “IoT-Based Rural Area Portable Diagnostic and Medication System,” designed to address these healthcare disparities by harnessing the capabilities of Internet of Things (IoT) technology. This system leverages IoT sensors and devices to enable remote health monitoring and diagnostics for residents of rural and underserved communities. It comprises a network of portable medical devices, such as wearable sensors, diagnostic tools, and medication dispensers, seamlessly connected through MQTT protocol to a centralized healthcare platform. Through this system, healthcare providers can remotely monitor patients’ vital signs, track disease progression, and deliver timely medical interventions. Moreover, the system ensures the secure and confidential exchange of sensitive medical data and enhances transparency between physicians and patients through the utilization of blockchain technology.

Voting is a fundamental right of citizens in a democratic country and crucial for any thriving democracy. Reliable voting systems are essential for free and fair elections in the modern era. Biometric Electronic Voting Machines (EVM) address many issues with paper ballot systems, but their closed-source nature undermines voter trust. Traditional election systems are also vulnerable to cyberattacks. This paper proposes a hybrid blockchain-based e-voting system (PQHAC-Bchain) to address the limitations of conventional e-voting systems and ensure a secure, auditable, tamper-proof, transparent, and privacy-preserving voting process. A scripting system for the Proposed blockchain facilitates a limited set of predefined operations for each layer, helping authoritative figures manage the election securely. This research uses Deepface face recognition and facial attribute analysis framework to confirm voter identity and prevent fraudulent voting. This research proposes a token-based approach to ensure secure and transparent voter identification while simultaneously preventing instances of double voting. This proposed blockchain system includes Post-quantum cryptography to protect against attacks from quantum computers. Through the integration of these technologies, an E-voting system that is both secure and transparent has been proposed in this paper. This system provides voters with the assurance that their votes are being accurately counted while safeguarding their privacy.

Bankruptcy is a lawful procedure when an organization is not able to repay its financial obligations to its investors. In this situation, one may lose their valuable possessions and disrupt regular industrial activities while paying off debts. This aforementioned situation has a significant impact on the financial stability of both individuals and organizations across the globe. Several works happened where different machine learning models were used to predict bankruptcy situations of different industries. In this work, we propose an Automated Machine Learning (AutoML)-based model that predicts the bankruptcy of different organizations more efficiently. First, the primary bankruptcy dataset was gathered from a public repository and preprocessed it for further processing. This dataset was balanced by oversampling with the Synthetic Minority Oversampling TEchnique (SMOTE) and undersampling with Random UnderSampling (RUS) techniques. To conduct the AutoML process, we employed the Tree-based Pipeline Optimization Tool (TPOT) framework in the primary, balanced oversampling, and undersampling datasets. After evaluating the classification outcomes of the primary and its balanced dataset, the best pipeline of Gradient Boosting in TPOTClassifier showed the best accuracy of 99.28% for the SMOTE oversampled dataset to predict bankruptcy more appropriately. This method also reduces many limitations of existing models and can be also used as a complementary tool in different machine learning applications.

The financial sector places significant emphasis on the detection of credit card fraud, and machine learning techniques have emerged as a promising solution. Nevertheless, questions regarding the interpretability of machine learning models and their resilience against adversarial attacks within the realm of credit card fraud detection persist. This study aims to bridge these gaps and underscore their pivotal role in advancing the field. It involves a comprehensive analysis of a credit card transaction dataset and the development of multiple machine learning models for fraud detection. Python scripts leverage well-known libraries such as Pandas, NumPy, Seaborn, Scikit-Learn, TensorFlow, and Matplotlib for data processing, visualization, and machine learning tasks. For fraud detection, the dataset is partitioned into training and testing sets, and three machine learning models, namely XGBoost, Random Forest, and Support Vector Machine (SVM), are applied. Performance metrics, encompassing accuracy, sensitivity (recall), F1-score, Matthews Correlation Coefficient (MCC), Balanced Classification Rate (BCR), and confusion matrices, are computed and presented. Additionally, a Long Short-Term Memory (LSTM) neural network is deployed for sequence-based fraud detection. This research offers a comprehensive roadmap for analyzing credit card transaction data and constructing machine learning models to detect fraudulent activities, underscoring the importance of data exploration and model evaluation in fraud detection systems. It emphasizes the significance of feature engineering in enhancing the effectiveness of fraud detection models and equips decision-makers with the ability to make informed choices based on model predictions.

Implementing the principles of inclusive education in higher education poses challenges, originally being developed for disabled students. As more students with disabilities successfully complete early schooling, the need for inclusive practices in higher education has grown. This research aims to share insights into inclusive practices within higher education, organized into three sections: describing the current situation of inclusive education for students with disabilities, reviewing literature on existing IoT technologies assisting students with disabilities in higher education, and discussing the necessary policies, strategies, processes, and actions to create a truly inclusive university setting that ensures success for all students. Disabled students often encounter additional learning difficulties due to physical or mental disabilities. This study reviews IoT technology-supported special education research, considering dimensions such as learning devices, strategies, domains, research issues, subjects, levels of disabilities, and learning environments. Results show recent studies, with a growing diversity in learning devices and applications. Data collected on UN-recognized countries, categorized over the economic condition, along with information on disabled populations, reveals disparities in opportunities for higher education. Developed countries, equipped with more resources and technologies, tend to have more inclusive educational systems. Conversely, there is a lack of data on disabled students in poor countries and their educational circumstances. The article highlights the use of IoT technologies in different countries, providing a breakdown of computing techniques and comparisons to determine suitable options demographically.

This paper explores the advantages of employing a selective relay selection method compared to a random relay selection approach in millimeter-wave (mmWave) networks assisted by amplify-and-forward (AF) relays, featuring two source-destination pairs and two relays. The investigation includes an analysis of the outage probability associated with the end-to-end signal-to-noise ratio (SNR) for users under both relay selection strategies in a log-normal shadowing channel. The proposed relay selection technique has been incorporated into relay-assisted mmWave networks, considering the shadowing effects caused by blockages. Simulation results demonstrate that the selective relay assignment outperforms random relay selection.

Segmentation of leaves is an initial step in image-based plant phenotyping, which has received more attention recently. Leaf segmentation shows features of the leaf and growth stages at the leaf level. Due to the significant overlapping between leaves and fluctuating environmental circumstances, such as intensity change and blur due to wind, segmentation of plant tissues like leaves is a most important issue in plant phenotyping. The leaf segmentation work has more complicated problems, such as leaf surface, genotypes, length, structure, and thickness variability. This study proposes an ultra-modern deep gaining knowledge of architectures: UUNet++, a convolutional neural network for advance segmentation. A vital assessment is done using Computer Vision for Plant Phenotyping dataset (CVPPP). The plant phenotyping is performed on tobacco and Arabidopsis leaf of RGB images from plant village dataset. The proposed model needs less space as compared to the main UNet architecture. In contrast to typical deep learning image sets, this dataset contains a limited number of samples. Despite that, we achieve exceptional results in leaf segmentation, specifically in distinguishing the interior of leaves from the binary segmentation of the entire leaf. Furthermore, studies are necessary to accurately quantify the number of leaves. Without reducing performance measures, the disease classification inferences time is extremely efficient and accurate.