Computer Science and Mathematics

Sort by

Article
Computer Science and Mathematics
Computer Vision and Graphics

Gregory Yu,

Aaron Collins,

Ian Butler

Abstract: Medical image segmentation systems, despite their recent sophistication, often face substantial performance degradation when exposed to unseen imaging environments—caused by differences in scanner types, acquisition protocols, or rare pathological conditions. To address this crucial issue, we introduce \textbf{MedSeg-Adapt}, a novel framework that enables \textit{clinical query-guided adaptive medical image segmentation}. MedSeg-Adapt features an autonomous generative data augmentation module that dynamically synthesizes environment-specific and clinically diverse training data using advanced medical image Diffusion models in combination with large language models (LLMs). This module automatically generates realistic image variants, natural language clinical queries, and pseudo-annotations—without requiring new reinforcement learning policies or manual labeling. In addition, we establish \textbf{MedScanDiff}, a new benchmark comprising five challenging medical imaging environments: Higher-resolution CT, Low-dose CT, Varying-field MRI, Specific Pathology Variant, and Pediatric Imaging. Extensive experiments demonstrate that fine-tuning state-of-the-art models such as MedSeg-Net, VMed-LLM, and UniMedSeg on MedSeg-Adapt-generated data significantly enhances robustness and segmentation accuracy across unseen settings, achieving an improvement of Dice Similarity Coefficient (DSC). MedSeg-Adapt thus provides a practical and effective pathway toward self-adaptive, clinically grounded medical image segmentation.
Article
Computer Science and Mathematics
Computer Science

Rao Mikkilineni,

Max Michaels

Abstract: Contemporary physics explains matter and energy with extraordinary success yet offers no principled account of how knowledge and mind arise and act causally in nature. We propose a Physics of Mindful Knowledge (PMK) in which knowledge-bearing constraints are recognized as organizing principles that stabilize metastable coherence and guide system dynamics. PMK integrates three pillars: (i) Burgin’s General Theory of Information (GTI), which formalizes information as a triadic relation (carrier–content–recipient), avoiding reification; (ii) Fold Theory, modeling the emergence and top‑down efficacy of coherent structures; and (iii) Deutsch’s epistemology, framing knowledge as “hard‑to‑vary explanations” and a constructor-level causal resource, using the Burgin–Mikkilineni Thesis (BMT), which reconceives computation as structural evolution and provides an engineerable path to oracle-like behavior with explanatory, predictive structure and teleonomy. We contrast PMK with microphysical (substrate‑dependent) and beyond‑physical (metaphysical) approaches and derive falsifiable predictions and measurable signatures—including Recovery Energy, Entropy per Useful Work, and Coherence Depth. We translate these into engineering protocols using structural machines (Mindful Machine Architecture) and outline comparative tests versus state‑of‑the‑art algorithmic systems under structural, functional, and epistemic perturbations. By embedding epistemic constraints into a naturalistic, computable framework, PMK provides a unified, testable bridge across physics, biology, and intelligent systems, advancing a practical science of knowledge, and mind. Physics of Mindful Knowledge (PMK) studies how matter–energy systems instantiate, transform, and use knowledge in mindful ways—meaning they monitor, evaluate, and adjust their own operations to preserve identity, purpose, and coherence.

Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Alex Mirugwe,

Arthur G. Fitzmaurice,

Alice Namale,

Evelyn Akello,

Simon Muhumuza,

Milton Kaye,

Samuel Lubwama,

Jonathan Mpango,

Paul Katongole,

Solomon Ssevvume

+4 authors

Abstract: Background: Duplicate patient records pose a significant challenge to healthcare registries and electronic medical record (EMR) systems in Uganda, primarily due to the absence of a national unique patient identifier. These duplicates lead to fragmented patient care, misallocation of resources, and inaccuracies in data reporting, which hinder effective monitoring of disease progression, disrupt continuity of care, and complicate efforts to track patient outcomes. Objective: To evaluate the performance of three classification algorithms in identifying duplicate records of people living with HIV (PLHIV) and to determine a combination of variables that can uniquely identify a PLHIV. Methods: The study used a six-step deduplication process involving dataset extraction, preprocessing, indexing, comparison, classification, and performance evaluation. Records of PLHIV who were active in care between June and November 2022 were extracted from the UgandaEMR system - an EMR installed at 15 public health facilities in six districts in the Rwenzori Region. The dataset included demographic variables, i.e., first name, middle name, last name, sex, age, date of birth, address, and phone number. Three classification algorithms were used to classify the client scores into matches, potential matches, and non-matches, namely i) a threshold-based algorithm, ii) a weighted average score-based algorithm, and iii) a decision tree. Due to the absence of a labeled dataset, the decision tree was trained on data labeled using the two rule-based methods and evaluated on a synthetic reference dataset. Performance of the algorithms was evaluated using sensitivity, specificity, and F-score metrics. Results: A total of 44,717 records for PLHIV active in care in the Rwenzori region from June to November 2022 were extracted. The weighted average score-based algorithm identified 447 (5.8%) records as duplicates and 2996 (10%) as potential duplicates. The threshold-based algorithm identified 118 (0.5%) duplicates and flagged 8560 (21.0%) as potential duplicates. The weighted average score-based algorithm achieved the highest performance: sensitivity (99.0%), specificity (98.8%), and F-score (98.9%); followed by the threshold-based classification: sensitivity (95.3%), specificity (89.1%), and F-score (92.1%); and the decision tree algorithm sensitivity (92.3%), specificity (93.9%, and F-score (93.1%). Conclusions: The weighted average score-based algorithm achieved the best performance. Findings highlight that a combination of a few demographic variables can be employed to differentiate PLHIV. However, improving duplicate record detection at scale will require training these algorithms on a larger dataset that can generalize the PLHIV population in Uganda.
Article
Computer Science and Mathematics
Information Systems

Seonghyeon Gong,

Jake Cho,

Kyuwon Ken Choi

Abstract: Provenance-based Intrusion Detection Systems (IDS) model the causal relationships between security events through a provenance graph and learn contextual information to detect Advanced Persistent Threats (APTs) effectively. However, existing provenance graph representation methods fail to fully reflect the characteristics of security domain data and the semantic information embedded in system logs, resulting in limitations in learning efficiency and detection accuracy. This paper proposes a provenance representation method that effectively captures security context from system log data. The proposed method improves the performance of provenance-based IDS by combining (1) a provenance graph construction technique that transforms meaningful string attributes—such as command lines, process names, and file paths—into vector representations to extract semantic information in the security context, (2) a hybrid time-position embedding technique for capturing causal relationships between events, and (3) an iterative refinement learning strategy tailored to the characteristics of system log data. Experimental results using the DARPA Transparent Computing Engagement 3 (E3) benchmark dataset for APT detection demonstrate that our method achieves improved accuracy compared to existing approaches while significantly accelerating convergence during iterative training. These results suggest that the proposed embedding technique can more effectively capture abnormal temporal patterns, such as long dwell times characteristic of APT attacks.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Aisha Sir Elkhatem,

Seref Naci Engin,

Yerbol Ospanov,

Aizhan Erulanova

Abstract: Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) remains challenging due to speckle noise, aspect-angle variation, and the loss of fine scattering cues in conven-tional deep-learning pipelines. Spatial-domain CNNs primarily extract geometric struc-ture but overlook high-frequency information critical for distinguishing small or spectrally similar targets, while frequency-only methods such as FFTNet fail to leverage spatial con-text and multi-scale spectral variation. To address these limitations, this study proposes the Multi-Scale Spectrum Pyramid Network (MSP-Net), which decomposes SAR images into low-, mid-, and high-frequency components via two-dimensional Fourier transforms with band-pass filtering and processes each band through dual convolutional branches equipped with predefined and learnable spectral filters. The resulting features are fused using attention-based, MLP-based, or transformer-based integration mechanisms. Expe-riments on two MSTAR-based benchmark datasets (11-class and 8-class) demonstrate that MSP-Net substantially outperforms spatial-only CNNs and single-scale frequen-cy-domain models. In the 11-class setting, MSP-Net improves accuracy by 13–14% (up to 95%) and achieves near-perfect ROC separability (AUC ≈ 1.0) with reliable calibration (ECE < 0.02). On the reduced 8-class dataset, the best MSP-Net variant achieves 99.9% ac-curacy and consistent per-class F1-scores. Ablation studies confirm the critical role of multi-scale spectral decomposition and adaptive fusion in improving recognition of small and spectrally similar targets such as BMP2, BTR60, and BTR70. These results highlight the effectiveness of frequency-aware, multi-scale learning for robust and interpretable SAR ATR.
Article
Computer Science and Mathematics
Algebra and Number Theory

James Hateley

Abstract: We develop an operator--theoretic framework for the Collatz map based on its backward transfer operator acting on weighted Banach spaces of arithmetic functions. The associated Dirichlet transforms form a holomorphic family that captures the complex--analytic evolution of iterates and admits a decomposition into a zeta--type pole at $s=1$ and a holomorphic remainder. Within a finer multiscale space adapted to the Collatz preimage tree, we establish a Lasota--Yorke inequality with an explicit contraction constant $\lambda<1$, giving quasi--compactness and a spectral gap at the dominant eigenvalue. The resulting invariant density is strictly positive and exhibits a $c/n$ decay profile. We formulate a general criterion showing that, under a verified quasi--compactness hypothesis with isolated eigenvalue $1$, the forward dynamics admit no infinite trajectories. The framework provides a coherent spectral perspective on the Collatz operator and suggests a broader analytic approach to arithmetic dynamical systems.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Laith H. Baniata,

Ashraf ALDabbas,

Jaffar M. Atwan,

Hussein Alahmer,

Basil Elmasri,

Chayut Bunterngchit

Abstract: Wireless sensor networks (WSNs) are increasingly being used in mission-critical infrastructures. In such applications, they are evaluated on the risk of cyber intrusions that can target the already constrained resources. The traditional intrusion detection systems (IDS) in WSNs are based on machine learning techniques. Such models fail to capture the nonlinear, temporal, and topological dependencies across the network nodes. Consequently, they cause degradation in the detection accuracy and poor adaptability against evolving threats. To overcome these limitations, this study introduced a hybrid deep learning-based IDS that integrated multi-scale convolutional feature extraction, dualstage attention fusion, and graph convolutional reasoning. In addition, bidirectional long short-term memory components are embedded into the unified framework. The proposed architecture captures the hierarchical spatial-temporal correlations in the traffic patterns. This allows making a precise discrimination between the normal and attack behaviors across several intrusion classes. The model has been evaluated on the benchmarking public available dataset and found to attain a higher classification capability in the multiclass scenarios. The model has further been found to outperform the conventional models focusing on the IDS frameworks. In addition, the proposed design is aimed at retaining suitable computational efficiency, which is suitable for edge and distributed deployments. This makes it an effective solution for the next-generation WSN cybersecurity. The overall findings have focused on combining topology-aware learning with multi-branch attention mechanisms for offering a balanced trade-off between interpretability, accuracy, and deployment efficiency for the resource-constrained WSN networks.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Mohammad Khalaf Rahim Al-juaifari

Abstract: Water scarcity poses a critical challenge to Iraqi agriculture, threatening food security and economic stability. This study develops an AI-driven precision irrigation framework for Iraq using real climate data from World Bank (2018-2023) and agricultural statistics from FAO. By integrating MODIS vegetation patterns with climate variables, we trained a Random Forest model (R² = 0.946) to optimize irrigation scheduling. Proposed analysis demonstrates that AI-driven irrigation can achieve 60% water savings compared to traditional methods while improving water use efficiency by 200%. The model identifies temperature (r=0.716) and NDVI (r=-0.713) as primary drivers of crop water stress, enabling precise irrigation timing during critical May-July periods. Economic analysis reveals potential annual benefits of $245 million through reduced water costs and maintained crop yields. This research provides a scalable framework for sustainable water management in arid regions, offering Iraq-specific solutions to address worsening water scarcity while maintaining agricultural productivity. The methodology demonstrates how AI can transform traditional agriculture using readily available satellite and climate data, with implications for water-stressed regions globally.
Article
Computer Science and Mathematics
Computer Vision and Graphics

Kaiqi Chen

Abstract: Video generation has rapidly advanced from early GAN-based systems to modern diffusion- and transformer-based models that deliver unprecedented photorealism and controllability. This survey synthesizes progress across foundational models (GAN, autoregressive, diffusion, masked modeling, and hybrids), information representations (spatiotemporal convolution, patch tokens, latent spaces), and generation schemes (decoupled, hierarchical, multi-staged, latent). We map applications in gaming, embodied AI, autonomous driving, education, filmmaking, and biomedicine, and analyze technical challenges in real-time generation, long-horizon consistency, physics fidelity, generalization, and multimodal reasoning. We also discuss governance and ethics, including misinformation, intellectual property, fairness, privacy, accountability, and environmental impact. Finally, we summarize evaluation methodologies (spatial, temporal, and human-centered metrics) and highlight future directions for efficient, controllable, and trustworthy video generation.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Zaid Khalid,

Noor Ul Amin

Abstract: Analyzing smartphone usage and behavior across different demographics, this study examines both self-reported and objective data from a dataset of 100 participants. Key variables such as total app usage, daily screen time, social media engagement, age, and gender were analyzed. The analysis reveals that social media contributes a significant portion of daily screen time, especially among the younger generation. The link between productivity and social media usage was also explored, revealing distinct behavioral patterns. Limitations of the study include a reliance on self-reported data and the static nature of the dataset. Future research should focus on longitudinal data to better understand the long-term effects of smartphone usage.
Article
Computer Science and Mathematics
Geometry and Topology

Edward Bormashenko

Abstract: We develop a topological-combinatorial framework applying classical Ramsey theory to systems of arcs connecting points on Jordan curves and their higher-dimensional analogues. A Jordan curve Λ partitions the plane into interior and exterior regions, enabling a canonical two-coloring of every arc connecting points on Λ according to whether its interior lies in Int(Λ) or Ext(Λ). Using this intrinsic coloring, we prove that any configuration of six points on Λ necessarily contains a monochromatic triangle, and that this property is invariant under all homeomorphisms of the plane. Extending the construction by including arcs lying on Λ itself yields a natural three-coloring, from which the classical value R(3,3.3)=17 guarantees the appearance of monochromatic triangles for sufficiently large point sets. For infinite point sets on Λ, the infinite Ramsey theorem ensures the existence of infinite monochromatic cliques, which we likewise show to be preserved under arbitrary topological deformations. The framework extends to Jordan surfaces and Jordan–Brouwer hypersurfaces in higher dimensions, where interior, exterior, and boundary regions again generate canonical colorings and Ramsey-type constraints. These results reveal a general principle: the separation properties of codimension-one topological boundaries induce universal combinatorial structures - such as monochromatic triangles and infinite monochromatic subsets - that are stable under continuous deformations. The approach offers new links between geometric topology, extremal combinatorics, and the analysis of constrained networks and interfaces.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Bernhard G Humm

Abstract: This article presents an industry-ready ontology for the machine learning domain, which is named “ML Ontology”. ML ontology is comprehensive, provides good performance and is extensible and adaptable. While based on lightweight modelling languages, ML ontology provides novel features including built-in queries and quality assurance, as well as sophisticated reasoning. Its industryreadiness is demonstrated by benchmarks as well as two use case implementations within a data science platform.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Zofia Feliksinska-Swierz,

Anetta Kępczyńska-Walczak,

Artur Wirowski

Abstract:

The objective of this paper is to develop a theoretical framework for the analysis of data exported from a Building Information Modeling (BIM) model through the application of Artificial Intelligence methods, serving as a foundation for risk assessment in construction projects. The purpose of this study is to investigate the potential of data mining techniques that function independently of biases introduced by predefined labelling. In recent years, a growing body of literature has examined the role of BIM technology in risk management. The most prevalent applications primarily rely on 3D visualization, which facilitates the identification and deeper understanding of potential issues related to design coordination and site safety. A significant contribution in this regard comes from built-in software features that enable automated clash detection and rule-based checking. Another dimension frequently associated with BIM in the context of risk management is 4D modeling, which incorporates construction sequencing to help mitigate risks related to buildability, scheduling, and subcontractor coordination. Based on a review of the relevant literature, this paper first presents a list of risk factors that can potentially be analysed using data extracted from BIM models, followed by an outline of a proposed method for further analysis employing machine learning techniques.

Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Jolien Van Bossche,

Thibault Clercq,

Callum Hensley,

Rune Peeters

Abstract: Visual question answering~(VQA) fundamentally requires a model to interpret heterogeneous semantic cues in an image and align them with a natural-language query. Traditional approaches benefit from scene graph representations, yet they often suffer from severe imbalances when handling rich semantic structures, especially when reasoning demands simultaneous consideration of objects, relations, and fine-grained attributes. Existing models frequently overlook the subtle interactions among these three information streams, leading to faulty attribute inference or overlooked relational cues. Addressing these long-standing limitations calls for a more principled integration of all semantic constituents within a unified and expressive reasoning space. In this paper, we introduce \textbf{\textsc{TriUnity-GNN}}, a tri-modal fusion framework that redefines scene graph reasoning by jointly enhancing object-centric, relation-centric, and attribute-centric representations under a unified graph neural paradigm. Instead of treating scene graphs as monolithic structures, our approach restructures the given graph into two complementary modalities, an object-dominant perspective and a relation-dominant perspective, thereby enabling the model to capture multi-granular semantics that are typically under-explored. To further strengthen the expressivity of these representations, \textsc{TriUnity-GNN} integrates attribute cues through an explicit fusion design, significantly enlarging the impact of attribute signals that are otherwise marginalized in classic architectures. Moreover, we design a novel message-passing enhancement module that substantially increases cross-type semantic exchange among objects, relations, and attributes, ensuring that all three modalities collectively shape the final reasoning embedding. We perform comprehensive evaluations on benchmark datasets including GQA, VG, and motif-VG. Across all benchmarks, \textsc{TriUnity-GNN} consistently surpasses prior graph-based VQA systems by a clear margin, demonstrating robustness in handling both straightforward and semantically composite queries. The results verify that a tri-modal, explicitly balanced graph reasoning mechanism is crucial for improving interpretability and accuracy in challenging visual question answering scenarios.
Review
Computer Science and Mathematics
Computer Vision and Graphics

Md Iqbal Hossain,

Neeresh Kumar Perla,

Afia Sajeeda,

Siyu Xia,

Ming Shao

Abstract: In the rapidly advancing domain of artificial intelligence, Vision-Language Models (VLMs) have emerged as critical tools by synergizing visual and textual data processing to facilitate a multitude of applications including automated image captioning, accessibility enhancements, and intelligent responses to multimodal queries. This survey explores the evolving paradigm of Pre-training, Fine-tuning, and Inference that has notably enhanced the capabilities of VLMs, allowing them to perform effectively across various downstream tasks and even enable zero-shot predictions. Despite their advancements, VLMs are vulnerable to adversarial attacks, largely because of their reliance on large-scale, internet-sourced pre-training datasets. These attacks can significantly undermine the models' integrity by manipulating their input interpretations, posing severe security risks and eroding user trust. Our survey delves into the complexities of these adversarial threats, which range from single-modal to sophisticated multimodal strategies, highlighting the urgent need for robust defense mechanisms. We discuss innovative defense strategies that adapt model architectures, integrate adversarially robust training objectives, and employ fine-tuning techniques to counteract these vulnerabilities. This paper aims to provide a comprehensive overview of current challenges and future directions in the adversarial landscape of VLMs, emphasizing the importance of securing these models to ensure their safe integration into various real-world applications.
Article
Computer Science and Mathematics
Probability and Statistics

Zdeněk Kala

Abstract: A Hermite-based framework for reliability assessment within the limit state method is developed in this paper. Closed-form design quantiles under a four-moment Hermite density are derived by inserting the Gaussian design quantile into a calibrated cubic translation. Admissibility and implementation criteria are established, including a monotonicity bound, a positivity condition for the platykurtic branch, and a balanced Jacobian for the leptokurtic branch. Material data for the yield strength and ductility of structural steel are fitted using moment-matched Hermite models and validated through goodness-of-fit tests. A truss structure is then analysed to quantify how non-Gaussian input geometry influences structural resistance and its corresponding design value. Variance-based Sobol sensitivity analysis demonstrates that departures of the radius distribution towards negative skewness and higher kurtosis increase the first-order contribution of geometric variables and thicken the lower tail of the resistance distribution. Closed-form Hermite design resistances are shown to agree with numerical integration results and reveal systematic deviations from FORM estimates, which rely solely on the mean and standard deviation. Monte Carlo simulation studies confirm these trends and highlight the slow convergence of tail quantiles and higher-order moments. The proposed approach remains fully compatible in the Gaussian limit and offers a practical complement to EN 1990 verification procedures when skewness and kurtosis have a significant influence on design quantiles.
Article
Computer Science and Mathematics
Computational Mathematics

Jean Chien,

Lily Chuang,

Nail Tang,

Eric Lee

Abstract: Nanoimprint lithography (NIL) master fidelity is governed by coupled variations beginning with resist spin-coating, proceeding through electron-beam exposure, and culminate in anisotropic etch transfer. We present an integrated, physics-based simulation chain. First, it includes a spin-coating thickness model that combines Emslie–Meyerhofer scaling with a Bornside edge correction. Second, it couples an e-beam lithography (EBL) module in which column electrostatics and trajectory-derived spot size feed a hybrid Gaussian–Lorentzian proximity kernel, followed by development thresholds are modulated by local thickness. Finally, it passes the exposure results to a level-set reactive ion etching (RIE) model with angular anisotropy and aspect-ratio-dependent etching (ARDE). With isolated and dense design layouts as bounding conditions, pattern fidelity is quantified by NMSE, ΔCD, and LER. The coupled analysis indicates that a low single-nanometer spot-size window trades dimensional accuracy for edge continuity; that over-widening generates proximity-dominated bias and feature coalescence; and that ARDE-informed evolution reproduces inward critical dimension (CD) drift in narrow openings, consistent with transport limitation. Collectively, the simulation chain accounts for stage-to-stage propagation from spin-coating thickness variation and EBL proximity to ARDE-informed etch profiles, and provides OPC-aligned metrics as outputs. In practical, mask process correction (MPC) is necessary rather than optional: the simulator serves as the predictive model, metrology supplies updates, and constrained optimization sets dose, focus, and etch set-points under CD/LER constraints.
Concept Paper
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Feng Chen

Abstract: Large language models (LLMs) are usually developed and evaluated as solitary agents: a single, monolithic network trained on static corpora and queried one prompt at a time. This single-agent paradigm has produced impressive capabilities, yet it fundamentally mismatches the structure of many real-world problems in science, engineering, and governance, which are inherently multi-actor, iterative, and argumentative. In this Perspective, we argue that the next scaling frontier for LLMs is not simply “bigger models with more data”, but societies of models and tools designed as structured collective intelligences. We first outline why classical scaling laws, which relate performance primarily to parameter counts, token volume, and compute, are insufficient for tasks that require debate, division of labor, and long-horizon coordination. We then introduce a conceptual framework based on three interaction regimes—competition, collaboration, and coordination—and show how different task families naturally demand different regime designs, incentives, and communication protocols. Building on emerging multi-agent LLM systems in reasoning, code generation, and autonomous science, we sketch a research programmer for “multi-agent pretraining”, in which agents jointly learn not only language and world models, but also norms of discourse, peer review, and self-correction. We further discuss how multi-agent architectures reshape scaling laws, evaluation methodology, and safety: performance becomes a function not only of model size and data, but also of team composition, interaction topology, and institutional memory. Finally, we argue that carefully engineered artificial communities may approximate the epistemic dynamics of real scientific communities more faithfully than any single, static model, opening a path toward more robust, transparent, and controllable AI systems.
Article
Computer Science and Mathematics
Mathematical and Computational Biology

Natalya Maxutova,

Akmaral Kassymova,

Kuanysh Kadirkulov,

Aisulu Ismailova,

Gulkiz Zhidekulova,

Zhanar Azhibekova,

Jamalbek Tussupov,

Quvvatali Rakhimov,

Zhanat Kenzhebayeva

Abstract: This paper proposes an intelligent and explainable ensemble system for predicting as-partate aminotransferase (AST) levels based on routine biochemical and demographic data from the NHANES dataset. The framework integrates robust preprocessing, adaptive feature encoding, and multi-level ensemble learning within a nested cross-validation (5×3) structure to ensure reproducibility and prevent data leakage. Several regression mod-els—including Random Forest, XGBoost, CatBoost, and stacking ensembles—were sys-tematically compared using R², RMSE, MAE, and MAPE metrics. The results show that the Stacking v2 architecture, combining CatBoost, LightGBM, and Ridge meta-regression, achieves the highest predictive accuracy and stability. Explainable AI analysis using SHAP revealed key biochemical and lifestyle factors influencing AST variability. The pro-posed system provides a modular, interpretable, and reproducible foundation for deci-sion-support applications in intelligent healthcare analytics, aligning with the goals of applied system innovation.
Review
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Aiperi Zhenishova

Abstract: This report presents a thorough consideration of the nascent area of Human-Centered Explainable Artificial Intelligence (XAI), concentrating on the crucial task of ensuring that AI decisions are understandable and credible to human users. With the spread of AI across sensitive domains like healthcare, finance, and online retail, the need for clear and understandable explanations increases. The review considers different formats of explanation such as visual aids (saliency maps, textual summaries) and analysis challenges faced in the evaluation process. Main finding: The research interest has also radically changed since 2021 from focusing on purely technical approaches to more on human perception, interaction and trust. We combine results of 73 published papers that exist until the year of 2024 from empirical research and show that local post-hoc explanation (particularly feature importance methods [e.g., LIME, SHAP]) is the current focus of much of the literature but that inherently interpretable models are treated with relatively little attention. Despite the large pool of explanation techniques, there is a dearth of standardized metrics to evaluate interpretability, user confidence, and impact on decision making. This gap restricts comparability of evidences between studies and hampers efforts to bring about efficient and user-friendly AI explanations. The paper calls for structured frameworks as well as a harmonized protocol for analyzing explainability - specifying how explanatory explanations would lead to greater user trust, understanding and support in makingdecisions. Ultimately, a humane, rigorous approach towards evaluating AI systems is necessary to not only make these transparent but also make them really understandable on the part of the reader. The goal of this work, in turn, is to drive further exploration to more trustworthy, human-centered modes of explanation that will bridge that the chasm between the complexity of algorithms and human understanding.

of 602

Prerpints.org logo

Preprints.org is a free preprint server supported by MDPI in Basel, Switzerland.

Subscribe

Disclaimer

Terms of Use

Privacy Policy

Privacy Settings

© 2025 MDPI (Basel, Switzerland) unless otherwise stated