The global displacement of people due to conflict, persecution, and political instability has reached unprecedented levels, creating significant challenges for humanitarian organizations and host nations. Accurate prediction of refugee and asylum seeker flows is crucial for effective resource allocation and policymaking. This work presents a data-driven approach to predicting the scale of displacement events using a comprehensive United Nations High Commissioner for Refugees (UNHCR) dataset. We extensively evaluate a diverse range of machine learning models, from traditional ensembles to deep learning architectures, including MLP Regressor, TabNet, FT-Transformer, and DistillBERT. A two-level Stacking Regressor combines XGBoost, LightGBM, and AdaBoost as base learners with a Ridge Regressor metalearner trained on their out-of-fold predictions to enhance final predictive performance. The proposed Stacking Regressor model achieved a remarkable R -squared value of 0.989 on a single holdout test set. A more rigorous 5 -fold cross-validation was conducted to assess model generalizability and robustness. These results identified LightGBM as the superior model, achieving the highest average $\boldsymbol{R}^{\mathbf{2}}$ score of 0.88, demonstrating its reliability and stability against data variance. As a key methodological contribution, we employ a comprehensive Explainable AI (XAI) framework using both LIME and SHAP to interpret model predictions, bridging the critical gap between predictive accuracy and the transparency required for real-world adoption. This work demonstrates the potential of interpretable and robust machine learning to support proactive and evidence-based humanitarian action.

Recent advancements in natural language processing (NLP) have employed transformer-based models to classify mental health conditions using social media data. Nevertheless, these models often face challenges with complex feature dependencies, require substantial computational power, and may not always yield optimal results. In contrast, conventional machine learning models can attain similar or even better performance by utilizing effective feature engineering, balancing data, and optimizing hyperparameters. This study addresses an important gap by demonstrating that a properly optimized Decision Tree model can greatly improve classification accuracy, surpassing deep learning methods such as BERT and BiLSTM. By utilizing Term Frequency-Inverse Document Frequency (TF-IDF), extracting N-gram features, and applying the Synthetic Minority Over-sampling Technique (SMOTE) for balancing classes, we improve the accuracy of predictions. Our Decision Tree model shows a 16 % increase in accuracy compared to transformer-based models, demonstrating that a well-tuned conventional machine learning method can compete with or exceed deep learning methods which have more computational cost.

Sudden cardiac arrest is a significant global health concern; however, triage of cardiac events typically occurs only in clinical settings, often when the patient is already experiencing a cardiac episode. Effective and efficient at-home cardiovascular virtual health monitoring is essential for early intervention, potentially preventing sudden fatalities. A highly accurate machine learning model combined with a Veroboardintegrated cardiac care kit has been developed in this work, which addresses the high mortality rate of sudden cardiac arrest through home-based monitoring. This device integrates three critical parameters-ECG, blood pressure, and heart rate-into a machine-learning model to analyze real-time cardiovascular health data. A combined dataset of$\mathbf{1, 6 9 0}$cardiac patients from the UC Irvine Machine Learning Repository is used for model training and testing, encompassing 11 health features, including critical cardiovascular indicators such as fasting blood sugar, ECG results, exercise-induced angina, and ST slope. While the dataset includes individuals across different age groups, it primarily focuses on individuals aged 40 to 60. Min-max scaling for continuous features and one-hot encoding for categorical features have been applied in the dataset preprocessing stage. A Stacking classifier is implemented, using Decision Tree, Random Forest, and Gradient Boost classifiers as base estimators, with KNN as the final meta estimator. The applied Stacking ensemble model achieves an accuracy of 95.6% and an F1 score of 95.9%. This proposed device ensures a user-friendly interface and high accuracy, making it suitable as a household monitoring tool to reduce fatalities from unanticipated cardiac events.

Diabetic retinopathy (DR) is a leading cause of vision impairment, which significantly impacts working-class populations, necessitating accurate and early diagnosis for effective treatment. Traditional DR classification relies on Convolutional Neural Network (CNN)-based models and extensive preprocessing. In this work, we propose a novel approach leveraging pre-trained models for feature extraction, followed by Graph Convolutional Networks (GCNs) for refined embedding representation. The extracted feature vectors are structured as a graph, where GCN enhances embeddings before classification. The proposed model incorporates quality assessment by predicting a confidence score through a dedicated fully connected layer, trained to align with ground truth quality using binary cross-entropy loss. Uncertainty estimation is achieved by calculating the variance across multiple stochastic passes, providing a measure of the model's prediction reliability. We evaluate the proposed DR detection approach on APTOS2019, Messidor-2, and EyePACS datasets, achieving superior performance over state-of-the-art methods. Using MobileViT as the main feature extractor, we reached a remarkable 98.45% accuracy, 98.45% F1-Score, and 98.06% Kappa on the APTOS2019 dataset. The DenseNet-169 proved to be the best backbone of the pipeline for the Messidor-2 dataset, with an accuracy of 94.90%, F1-Score of 94.87%, and Kappa of 93.63%. Additionally, for external validation, the model demonstrated strong generalization capability on the EyePACS dataset, where DenseNet-169 achieved 97.38% accuracy, 97.37% F1-Score, and 96.72% Kappa, while MobileViT obtained 96.02% accuracy, 96.02% F1-Score, and 95.03% Kappa. Our innovative architecture incorporates uncertainty estimation and quality assessment techniques, enabling accurate confidence scores and enhancing the model's reliability in clinical environments. Furthermore, to strengthen interpretability and facilitate clinical validation, Grad-CAM heatmaps were employed to demonstrate the significance of different input regions on the model's predictions.

Diabetic retinopathy (DR) is a critical complication of diabetes and is characterized by damage to retinal blood vessels. Without early diagnosis and treatment, it can lead to vision loss. The objective of this study is to investigate diverse methodologies to address the class imbalance issue in the retinal fundus image dataset and apply transformer-based knowledge distillation (KD) for efficient DR detection. Various deep-learning models have been implemented to classify RGB fundus images into five distinct classes of DR. The retinal fundus images of the employed APTOS dataset were classified using the following models: ResNet34, DenseNet121, GoogLeNet MobileViTv2, DeiT3, and KD. Initial training on the imbalanced APTOS dataset showed promising results, but to explore a more robust approach, two data balancing approaches were used, namely cost-sensitive learning and data augmentation. It was found that MobileViTv2 achieved the best results with the APTOS cost-sensitive learning approach, achieving a kappa score of 0.94 and a macro F1 score of 0.73. Furthermore, the KD technique with MobileViTv2 as a teacher and GoogLeNet as a student model achieved a WKS of 0.90 with significantly fewer parameters than the teacher model. Finally, the automatic predictions of the deep learning models are interpreted using the SHAP explainable AI framework. These findings suggest that the cost-sensitive approach and lightweight KD model can substantially enhance the performance of the implemented deep learning models for DR detection in memory-constrained healthcare devices.

Mosquito-related diseases pose a significant threat to global public health, necessitating efficient and accurate mosquito classification for effective surveillance and control. This work presents an innovative approach to mosquito classification by leveraging state-of-the-art vision transformers and open-set learning techniques. A novel framework has been introduced that integrates Transformer-based deep learning models with comprehensive data augmentation and preprocessing methods, enabling robust and precise identification of ten mosquito species. The Swin Transformer model achieves the best performance for traditional closed-set learning with 99.60% accuracy and 0.996 F1 score. The lightweight MobileViT technique attains an almost equivalent accuracy of 98.90% with significantly reduced parameters and model complexities. Next, the applied deep learning models' adaptability and generalizability in a static environment have been enhanced by using new classes of data samples during the inference stage that have not been included in the training set. The proposed framework's ability to handle unseen classes like insects similar to mosquitoes, even humans, through open-set learning further enhances its practical applicability employing the OpenMax technique and Weibull distribution. The traditional CNN model, Xception, outperforms the latest transformer with higher accuracy and F1 score for open-set learning. The study's findings highlight the transformative potential of advanced deep-learning architectures in entomology, providing a strong groundwork for future research and development in mosquito surveillance and vector control. The implications of this work extend beyond mosquito classification, offering valuable insights for broader ecological and environmental monitoring applications.

Bidomain or monodomain model has frequently been used to study electrical activities in the cardiac tissue. The finite difference method with second order accuracy is commonly used to approximate the spatial derivatives of the governing equations numerically. In this work, a higher order finite difference scheme has been implemented to solve the nonlinear monodomain equation. The unknown transmembrane potential is expanded in terms of Lagrangian interpolating polynomials. Differentiation of the polynomial expansion then gives the finite difference approximation, and the order of the approximation is varied by changing the order of the polynomial. First and second order semi-implicit (implicit-explicit) time-stepping techniques have been used to approximate the temporal derivatives. An explicit finite difference scheme with 512×512×512 nodes and 0.1 μs time step is used as the benchmark for error calculation. As the order of the finite difference approximation is increased, the error potential reduces. However, there is less noticeable error improvement after the ninth order approximation. The use of an operator splitting method as well as a protective zone scheme further improves the performance of the semi-implicit scheme. Numerical results demonstrate that the error can be reduced by a factor of 2.6, while the CPU time can be reduced by a factor of 14.1.