Humans can carry various diseases, some of which are poorly understood and lack comprehensive solutions. Such a disease can exists in human eye that can affect one or both eyes is diabetic retinopathy (DR) which can impair function, vision, and eventually result in permanent blindness. It is one of those complex complexities. Therefore, early detection of DR can significantly reduce the risk of vision impairment by appropriate treatment and necessary precautions. The primary aim of this study is to leverage cutting-edge models trained on diverse image datasets and propose a CNN model that demonstrates comparable performance. Specifically, we employ transfer learning models such as DenseNet121, Xception, Resnet50, VGG16, VGG19, and InceptionV3, and machine learning models such as SVM, and neural network models like (RNN) for binary and multi-class classification. It has been shown that the proposed approach of multi-label classification with softmax functions and categorical cross-entropy works more effectively, yielding perfect accuracy, precision, and recall values. In particular, Xception achieved an impressive 82% accuracy among all the transfer learning models, setting a new benchmark for the dataset used. However, our proposed CNN model shows superior performance, achieving an accuracy of 95.27% on this dataset, surpassing the state-of-the-art Xception model. Moreover, for single-label (binary classifications), our proposed model achieved perfect accuracy as well. Through exploration of these advances, our objective is to provide a comprehensive overview of the leading methods for the early detection of DR. The aim is to discuss the challenges associated with these methods and highlight potential enhancements. In essence, this paper provides a high-level perspective on the integration of deep learning techniques and machine learning models, coupled with explainable artificial intelligence (XAI) and gradient-weighted class activation mapping (Grad-CAM). We present insights into their respective accuracy and the challenges they face. We anticipate that these insights will prove valuable to researchers and practitioners in the field. Our ambition is that this in-depth study and comparison of models will inform and inspire future research endeavors, ultimately leading to enhanced disease detection in medical imaging and thereby assisting healthcare professionals.

Among the major occupational sectors around the world, agriculture has the highest level of involvement. Every year, this sector faces a substantial loss in production and profit due to a large number of diseases in crops and plants. If those diseases are not detected early and taken measures for prevention, it can bring about a devastating result that can financially burden agriculture personnel. Traditional methods of detecting diseases in plants and crops offer high accuracy. However, the procedure is time-consuming, which might be insidious in most cases. Crop diseases need to be detected and cured as soon as possible as most diseases are highly contagious among crops and plants. In this paper, we have used the transfer learning approach with three pre-trained models: EfficientNetV2L, MobileNetV2, and ResNet152V2, to detect various plant diseases. We have proposed a framework to detect 38 types of leaf diseases in 14 different plants, compared the three pre-trained models based on various quantitative performance evaluation parameters, and demonstrated that EfficientNetV2L performed best with 99.63% accuracy. In the end, Explainable Artificial Intelligence (XAI) technique: LIME has been implemented in our model to understand the insight view of the model EfficeintNetV2L's for such prediction. It is used to make our model's predictions more reliable and gives a clear explanation about the reason of such decision.

Vitamin D deficiency is becoming a global public health concern, particularly in medical centers, with serious consequences for disease severity, mortality, and morbidity. Traditional diagnostic methodologies struggle with cost and are time-consuming. This drives a paradigm change toward the use of automated process for improved predicted accuracy and cost-effectiveness in diagnosing Vitamin D deficiency. In addressing the global challenge of Vitamin D deficiency, this study introduces an ensemble model that synergistically combines LightGBM and CatBoost algorithms, marking a significant leap in diagnostic methodologies. By leveraging the strengths of these advanced machine learning techniques, our approach achieves an impressive 96% accuracy on the Vitamin D Deficiency (VDD) Dataset, demonstrating substantial improvement over traditional diagnostic methods. The integration of SnAP (Shapley Additive explanations) for interpretability further enhances the utility of our model, providing clear insights into the impact of individual features on the severity predictions of Vitamin D deficiency. This revised abstract concisely encapsulates the research objectives, innovative methodology, and critical findings, underscoring the potential to revolutionize diagnostic efficiency and accuracy in the medical domain.

In a world with 7,000+ languages, Papua New Guinea leads with 839. Indonesia has 706, China 301, and Arabic features numerous variations. Understanding emotional nuances in diverse languages poses a challenge, especially for non-native speakers, tourists, and influencers. Social media has emerged as a vital information source for tourism, influencing, and vlogging. This study aims to outline the use of sentiment analysis in social media platforms. It provides valuable insights for travelers, vloggers, foreigners, and influencers, aiding them in making informed decisions and ensuring safety and satisfaction. It compares emotion identification across nine languages and evaluates SVM, LSTM, BiLSTM models with two word embeddings. The goal is to classify tweets into Positive, Neutral, and Negative emotions. Our Custom Model consistently outperforms existing works. It achieves an impressive 90.01% accuracy with a dataset exceeding 270,399 sentences in multiple languages.

Object detection plays a pivotal role in enhancing security, surveillance, and automation. It enables timely threat identification, streamlines traffic management, and facilitates efficient resource allocation. By automating object recognition, it creates value through improved safety, productivity, and resource optimization in various domains, from smart cities to industrial settings. This paper explores modern object detection methods to develop a Real-Time Responsive CCTV Camera Model. As we transition to a 5G-connected world, smart devices are poised to manage our daily security. We focus on transforming conventional CCTV cameras into responsive, internet-connected security guards. By evaluating methods like HOG, Viola-Jones, R-CNN, SSD, and YOLO, we aim to select the most efficient algorithm. We found Yolov8 perform best among all the models based on accuracy 99.8% and FPS 40. Our research strives to create cost-effective, intelligent security systems, paving the way for automated alerts and a digitally secured future.

Despite progress in global healthcare systems, the utilization of domestic healthcare facilities remains limited in several regions, with a considerable proportion of people pursuing treatment overseas. This development highlights the necessity for systematic incorporation of patient-centered input, an essential element for enhancing accountability, transparency, and quality in local healthcare. Our research seeks to address this deficiency by establishing a system that collects and analyzes patient feedback to inform and improve healthcare policies and practices, particularly in areas with elevated demand for medical services. We offer an effective platform for viewing reviews from different hospitals, especially in places where people routinely visit for medical services. Therefore, we built our primary ”Dhaka Private Hospitals Review Dataset,” considering gathering and evaluating patient opinions methodically. We further employ transformer-based generative adversarial learning to evaluate sentiment analysis using knowledge distillation (KD) to boost model efficiency. Our proposed GANBERT architecture includes two optimized student models gated recurrent unit-based Contextualized BERT (GC-BERT) and LSTM-based Contextualized BERT (LC-BERT) with enhanced generators and discriminators. Our GC-BERT enhances execution time by 1.27% to 24.27%, while LC-BERT improves by 14.13% to 23.30%, showing superior advancements compared to other contemporary models. Each model with reductions ranging from 82.50% to 99.99% parameters, making them lightweight and efficient compared to other teacher models in the KD process. Instead of using contextual word representations which demand more space and complexity for reviewing patient feedback, we utilize the single static pretrained and low-dimensional word embedding space approach integrating student models.