Abstract—Popularity of Social media sites has been on the rise ever since its invention. Due to the anonymity the social media sites provide to the users, an influx of divergent behavior has also been increasing. Abusive language and offensive content are posted on these sites to ridicule others which creates a harmful environment for the people affected. Detecting these misconducts, especially for Bengali Language is still vague. Our paper tries to bridge the gap between the abusive content detection using Natural Language Processing with five different machine learning classifiers such as Decision Tree, Random Forest, Multinomial Naïve Bayes, Support Vector Classifier and Logistic Regression and give clarity on the factors that contribute to the prediction using Explainable AI. The dataset was preprocessed in multiple steps including stop words and punctuation removal, null and duplication value check and tokenization. The dataset was then processed to created TF-IDF representations with n-grams. Grid Search Cross Validation technique was used to find out optimal combination of hyperparameters for our machine learning models. Among all the models Logistic Regression performed the best with an accuracy of 85.57%. For transparency and clarification of the model’s detection methodologies Lime model was used. 

Keywords—Machine Learning, NLP, XAI, Bangla Comments, Online Abuse .

Abstract—In today's developing world, anxiety is a common mental disorder among university students. In this work, we predict anxiety in university students using a voting classifier. We have applied explainable artificial intelligence (XAI), to gain a better understanding of the machine learning model, using a Google Form, the dataset was gathered from several public, private, and national universities in Bangladesh. We have compared machine learning algorithms using 20 selected features and without feature selection. By using the concept of voting, we have created a new model. In order to create our final model, we selected the best three ML algorithms based on their accuracy. The voting classifier has the highest accuracy of 96%, while F1 and Recall scores are 96%, and Precision is 97%. The LIME and SHAP models are used to explain model predictions instead of a black-box machine learning model. The study determines anxiety levels by particular observations, allowing for transparency and comprehension. The goal is to create prediction models to detect at-risk individuals and root causes of anxiety among university students, consequently reducing detrimental consequences for academic performance and wellbeing. 

Keywords—Mental health, Anxiety, Machine Learning, XAI, LIME, SHAP, University Student, Bangladesh. 

Abstract— In the realm of fruit quality assessment, the demand for accurate identification methods has grown significantly. Traditional measures, such as physical and chemical analyses, prove impractical due to their timeconsuming nature and potential sample damage. To address this imperative need for precise and timely apple quality detection, this research explores the application of machine learning algorithms. Recognizing the drawbacks of traditional methods reliant on physical and chemical measures for predicting fruit maturity, non-destructive techniques, such as hyperspectral imaging and machine learning algorithms, are meticulously investigated. The study goes on to evaluate and compare the performance of various machine learning approaches, including Random Forest Classifier (RF), K-Nearest Neighbors Classifier (KNN), Support Vector Classifier (SVC), Extreme Gradient Boosting Classifier (XGB), Light Gradient Boosting Machine Classifier (LGBM), Artificial Neural Network (ANN), and Voting Classifier. Among these, the Artificial Neural Network emerges as the top performer, achieving an impressive accuracy rate of 92.87%. Following the thorough evaluation of the models, Explainable Artificial Intelligence (XAI) techniques, specifically LIME and SHAP, are employed. These techniques play a crucial role in interpreting and visualizing the intricate decision-making process of the models. The insights gained underscore the models' robustness while providing valuable information on the decision-making processes, contributing significantly to advancements in apple quality detection methodologies. 

Keywords— Apple quality, fruit classification, machine learning, ANN, XAI, LIME, SHAP 

Abstract— Paddy production is one of the significant branches of the agricultural sector that is vulnerable to being affected by a range of fungal, bacterial, and viral infections. These diseases can cause significant yield losses and reduce the quality of rice crops if not managed effectively. To address this issue, automation of paddy disease detection plays a vital role in enhancing productivity and promoting sustainable farming. In our study, we presented an innovative deep-learning approach for the automated detection of paddy diseases considering six distinct disease types, including Bacterial Blight, Blast, Brownspot, Leaf scaled, Sheath Blight, and Tungro along with healthy leaves. We utilized a composite dataset of 8526 images collected from Kaggle and Mendeley consisting of seven classes. Two deep learning models: Convolutional Neural Network (CNN) and ResNet-50 were applied in the process. The CNN model achieved an accuracy of 88% on the employed dataset while ResNet-50 exhibited superior performance of 97% accuracy. The model’s performance is evaluated using confusion matrices containing precision, recall, accuracy, and F1-score. The pinnacle of the research includes the deployment of the enhanced ResNet-50 model into a mobile application designed using Flutter and FastAPI. This application serves as a testament to the potential of real-time agricultural diagnostics, enabling users to capture images of paddy crops and receive instant disease predictions, signifying a transformative advancement in agricultural technology with timely insights to effectively manage crop health and optimize yields. 

Keywords—Paddy disease detection, Deep Learning, CNN, ResNet-50, Flutter, FastAPI, Agricultural 

Abstract— Depression, a pervasive mental disorder, significantly impacts individuals' daily lives, manifesting as persistent sadness, a negative disposition, lack of interest or pleasure, feelings of guilt or worthlessness, and sleep disturbances. Advances in psychological methodologies, including the use of questionnaires and the analysis of social media posts, have facilitated the detection of individuals affected by this mental health issue. While mental health is a priority in many developed countries, it remains primarily neglected in regions like Bangladesh. This neglect is evident in the lack of a standardized corpus for detecting depression in the area. In this paper, we present the annotated data collected by our crawlers and detail the processes undertaken to standardize this corpus. We also propose a novel model for depression analysis and detection, which integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. Our model achieves an accuracy of 90.11%, demonstrating its efficacy. Additionally, we compare the proposed model with various machine learning models, including SVM, Random Forest, Naïve Bayes, KNN, AdaBoost, CNN, LSTM, and Decision Tree with hyperparameter tuning, using several performance metrics to highlight the advantages of our approach. 

Keywords—Depression, Deep Learning, LSTM, Machine Learning 

Abstract— An essential component of the diagnostic and treatment process is identifying brain tumors early in their onslaught. Traditional approaches struggle with processing sequential data and face limitations in maintaining computational efficiency with large-scale data. The pre-trained model has the advantage of eliminating computational complexity and processing the data very well. This research intends to improve the overall accuracy and robustness of the brain tumor detection and classification system, which is an amalgamate of several CNN-based pre-trained models for classification and two well-known detection models, Mask RCNN and YOLOv8, for tumor detection. The ResNet50 model delivers remarkable results, with an accuracy of 98.21% and precision of 97.32%. On top of that, it has a recall of 97.09% and an F1 score of 97.01%. The detection approach successfully pinpoints the tumor and offers a confidence score for detected regions exceeding 0.8. It also makes a highly accurate distinction between positive and negative tumors. The findings showcase significant benefits for medical imaging, offering enhanced interpretable results through the use of Grad CAM as the optimal classifier. 

Keywords— Brain Tumors, Deep Learning, CNN, Mask RCNN, YOLOv8, Grad CAM, Medical Imaging 

Abstract— Glaucoma is a leading cause of irreversible blindness worldwide, necessitating early and accurate diagnosis to prevent vision loss. To ensure trustworthiness in the accuracy of diagnosis, interpretability, and understanding of risk factors associated with glaucoma are essential. This study proposes a comprehensive methodology for glaucoma classification using various deep learning architectures, including CNN, VGG16, VGG19, InceptionResNetV2, Xception, and Vision Transformers (ViTs) integrated with Explainability techniques. The dataset utilized for this research comprises retinal fundus images divided into training, validation, and testing sets. The approach involves applying various preprocessing techniques, like CLAHE, green channel conversion, and canny edge detection, to enhance feature extraction. Among the evaluated models, the ViT demonstrated superior performance, achieving an accuracy of 92% on the test set. This model's ability to dynamically adjust attention weights based on the input image contributes to its improved classification capabilities. To provide insights into the model’s decision-making process, Grad-Cam facilitates better treatment planning for clinicians.

Keywords— Glaucoma, Deep Learning, Explainable Artificial Intelligence (XAI), Vision Transformers, Grad CAM, CLAHE. 

Abstract— An essential component of the diagnostic and treatment process is identifying brain tumors early in their onslaught. Traditional approaches struggle with processing sequential data and face limitations in maintaining computational efficiency with large-scale data. The pre-trained model has the advantage of eliminating computational complexity and processing the data very well. This research intends to improve the overall accuracy and robustness of the brain tumor detection and classification system, which is an amalgamate of several CNN-based pre-trained models for classification and two well-known detection models, Mask RCNN and YOLOv8, for tumor detection. The ResNet50 model delivers remarkable results, with an accuracy of 98.21% and precision of 97.32%. On top of that, it has a recall of 97.09% and an F1 score of 97.01%. The detection approach successfully pinpoints the tumor and offers a confidence score for detected regions exceeding 0.8. It also makes a highly accurate distinction between positive and negative tumors. The findings showcase significant benefits for medical imaging, offering enhanced interpretable results through the use of Grad CAM as the optimal classifier. 

Keywords— Brain Tumors, Deep Learning, CNN, Mask RCNN, YOLOv8, Grad CAM, Medical Imaging 

Abstract— An excessive force can cause a bone fracture and result in problems such as bleeding, infection, hypoxia, interfragmentary strain, and loss of support. The manual procedure of identifying fractures from X-rays or CT scans mostly depends on orthopedic surgeons' expertise, which might be difficult in remote regions. Medical diagnoses in real time are becoming feasible with deep learning and machine learning development. The datasets used in this study were obtained from Kaggle. The first dataset, titled "Bone Fracture Detection: Computer Vision Project," included 4,148 labeled images and 10,581 X-ray images segmented into "fractured" and "not fractured" categories for bone fracture classification. This research aims to use deep learning models like CNN, ConvNeXt, ViT, MobileViT, VGG16, and VGG19 to detect bone fractures and characterize their severity effectively. YOLOV8 is utilized for fracture diagnosis, and image processing techniques improve noisy X-ray images obtained 0.995 with maPA50 and 0.991 with maPA50-95. In terms of classification accuracy, MobileViT showed 99%. This research aims to develop a reliable, automated system for detecting bone fractures facilitating early diagnosis and treatment.

Keywords— Bone Fracture Detection, Deep Learning, X-ray Images, YOLOv8, MobileViT, Medical Image Processing 

Abstract— An essential component of the diagnostic and treatment process is identifying brain tumors early in their onslaught. Traditional approaches struggle with processing sequential data and face limitations in maintaining computational efficiency with large-scale data. The pre-trained model has the advantage of eliminating computational complexity and processing the data very well. This research intends to improve the overall accuracy and robustness of the brain tumor detection and classification system, which is an amalgamate of several CNN-based pre-trained models for classification and two well-known detection models, Mask RCNN and YOLOv8, for tumor detection. The ResNet50 model delivers remarkable results, with an accuracy of 98.21% and precision of 97.32%. On top of that, it has a recall of 97.09% and an F1 score of 97.01%. The detection approach successfully pinpoints the tumor and offers a confidence score for detected regions exceeding 0.8. It also makes a highly accurate distinction between positive and negative tumors. The findings showcase significant benefits for medical imaging, offering enhanced interpretable results through the use of Grad CAM as the optimal classifier. 

Keywords— Brain Tumors, Deep Learning, CNN, Mask RCNN, YOLOv8, Grad CAM, Medical Imaging