Wasit Journal of Computer and Mathematics Science

The most accessible radiological technique for thoracic disease detection and
diagnosis uses chest X-ray imaging as its primary method. The deployment of automated chest Xray
analysis systems faces two main obstacles because of untrustworthy labels in large datasets
and unpredictable predictive confidence levels. The research proposes a hybrid system which
combines Vision Transformer (ViT) architecture with methods to handle noisy labels and produce
accurate probability estimates for multiple disease diagnosis in chest X-ray images. The system
trains on CheXpert and NIH ChestX-ray14 datasets while using Co-Teaching and DivideMix
noise-handling methods and self-supervised pretraining to enhance feature resistance against
supervision errors. The framework uses temperature scaling and Monte Carlo dropout as post-hoc
methods to enhance confidence reliability without compromising discriminative performance. The
system aims to reach performance levels that match or exceed traditional CNN and standard ViT
models in AUROC and mAP and F1 score metrics. The system reduces the effects of
untrustworthy labels while generating meaningful confidence scores which doctors can
understand. The model produces Grad-CAM++ explanations to assist doctors in understanding its
decision-making process. The hybrid system works to develop AI systems that deliver both
exact results and safe operational readiness for real-world chest X-ray decision support systems.

Iris recognition is a biometric technology widely used in applications including border control, voter
registration, and mobile authentication. However, variations in light, reflection, and noise in the image cause issues
with accurate recognition. This paper presents a two-stage iris recognition system. The first stage segmentations and
detection of the iris is performed using the HoughCircles algorithm, as well as inpainting the image to enhance image
quality and remove any shading or artifacts. The second stage uses novel multi-type neural convolutional network
that combines standard, spatial, and depthwise convolutions as part of its feature extraction and classification. The
proposed model is evaluated on the MULBv1 dataset to obtain 98.21% classification accuracy and a 98.16% F1-
score. The results demonstrate the model’s efficacy for use in real-world iris identification systems.

Feature selection is a crucial element of machine learning model design, mainly in sensitive areas
such as HR, finance, and healthcare. Evidence from the literature suggests that, although the current research
utilizes models with notable accuracy, fairly and interpretability are usually neither assessed nor the main focus of
the study. This study describes a feature selection method which incorporates Mutual Information for feature
relevance, while obtaining constraints for fairness with the Equal Opportunity Difference (EOD) metric. The goal of
this study is to maximize accuracy, fairness, and interpretability. The method was utilized for six methods on the
IBM HR Analytics dataset (Fair Logistic Regression, Explainable Boosting Machines (EBMs), XGBoost, Random
Forest, SVM, and KNN). Fair Logistic Regression, yielded the best results overall with 96% accuracy and little
fairness bias (EOD = 0.005). XGBoost and Random Forest also had predictive reliability, yet evidenced a disparity
in fairness measures. In addition, EBMs negotiated the accuracy values and have advantages of interpretability, thus
these models can be used in intervention areas where the transparency of the model/construction process is
mandated.

This paper addresses the inverse problem of reconstructing complete steady-state solutions for elliptic partial differential equations when boundary information is incomplete a situation common in electromagnetic, thermal, and geophysical modeling where full Dirichlet or Neumann data are not available. The objective is to develop a numerically stable and reproducible method that reconstructs the interior field and missing boundary trace from partial boundary measurements contaminated by noise. To that end, the inverse task is posed as a Tikhonov-regularized optimization problem in which the unknown Dirichlet trace on the unobserved boundary segment minimizes a least-squares discrepancy between forward-model predictions and boundary observations; the forward PDE is discretized by the finite element method and the gradient of the discrete cost is computed via the adjoint-state method, enabling efficient computation independent of the number of boundary parameters. The optimization uses a nonlinear conjugate-gradient scheme with automated Lcurve-
based on regularization selection. Numerical validation on synthetic test cases, including a unit-square Poisson problem and a circular-domain example, shows that the proposed approach recovers the interior field and the missing boundary trace with low RMSE and small relative L^2 errors for noise levels up to 5%, whereas unregularized reconstructions suffer severe noise amplification. These results demonstrate that the FEM + adjoint + H¹-Tikhonov methodology provides a practical and stable solution approach for elliptic inverse problems with partial boundary conditions, and that this approach scales to moderate resolution meshes and is potentially extendable for non-smooth targets, nonlinear PDEs, and realistic measurement models.