Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Faaz Khan Mohammad

Abstract: A simulated psychiatric patient must identify what a learner is asking about before it can decide whether to disclose a case detail. We evaluated this matching step on 512 AI-generated questions across four fictional psychiatric cases. The benchmark writer saw descriptions of the case memories but had no access to the matching rules. We reserved 248 questions for development and 264 for testing. On the test set, keyword matching recalled 35.1% of expected risk memories. Adding a small sentence encoder raised recall to 72.7%, while the share of questions about non-risk memories incorrectly matched to a risk memory rose from 6.1% to 15.2%. The hybrid matcher took a median of 2.1 ms per question on a laptop CPU. Adding hand-set heads raised recall to 77.9% and the false-hit rate to 21.2%. All methods fell short of the 95% recall target. The hybrid matcher scored 90.7% on the development set, where we also used missed questions to write probes. A separate attempt to maximize recall reached 100% on development data at a 36.9% false-hit rate. Tested in scripted conversations, that configuration sometimes prompted unasked disclosures. The results show why evaluation should report both recall and false hits on held-out questions.

Article
Engineering
Bioengineering

Aleksandra Świetlicka

,

Marcin Fojcik

,

Gunilla Kulla

,

Karol Perliński

Abstract: The growing demand for efficient patient monitoring, combined with the projected shortage of healthcare professionals, necessitates innovative solutions to support medical staff. This concept paper presents a framework for a contactless temperature monitoring system for hospitals and nursing homes, combining thermal imaging, RGB-based anatomical localization, and machine learning for continuous and personalized monitoring. The framework addresses key challenges related to temperature acquisition, environmental variability, individual physiological differences, data processing, and patient privacy. Preliminary experiments using thermal imaging and data from a single participant demonstrate the feasibility of representing contactless temperature measurements as temporal signals and detecting deviations from an individual temperature profile using an LSTM Autoencoder. The paper therefore identifies key technological challenges and research needs and proposes a roadmap toward robust, personalized, and clinically applicable contactless temperature monitoring.

Article
Engineering
Mechanical Engineering

Aman Kumari

Abstract: Viscoelastic surfactant (VES) gels are attractive fracturing fluids because they break down on their own when they meet oil or formation water, leaving little residue in the fracture. Their weak points are poor heat resistance and no filter-cake control. One way around this is to blend a zwitterionic surfactant with an anionic one, so that the mixture forms stronger self-assembled structures than either surfactant alone. This paper looks at two such blends, cocamidopropyl betaine (CAPB) with sodium oleate (NaOA) and CAPB with sodium dodecyl sulfate (SDS), each combined with iso-amyl alcohol as co-surfactant, clove oil, and water. Drawing on the experimental data of Baruah et al., we compare the critical micelle concentration, phase behavior, flow curves, temperature and pressure response, viscoelastic moduli, miscibility, and proppant settling of the two systems. We also add simple checks of our own: viscosity at pumping shear computed from the power-law fits, a Stokes-law ranking of the three proppants, and unit conversions to field practice. The CAPB + NaOA blend gels at lower surfactant content and is clearly stronger. With 0.1% NaOH and 0.1% SiO2 nanoparticles at 28% co-surfactant/surfactant, it kept its viscosity at 103 °C and 100 s−1 for 120 min at 2068–6205 kPa, thinned out in water but not in diesel, and held sand, intermediate-strength, and bauxite proppants up to 100 °C.

Article
Physical Sciences
Particle and Field Physics

Raghu Kulkarni

Abstract: Koide's relation, \(K = (m_e + m_\mu + m_\tau)/(\sqrt{m_e}+\sqrt{m_\mu}+\sqrt{m_\tau})^2 = 2/3\), holds for the charged-lepton pole masses to about \(10^{-5}\) and has no accepted explanation. It is equivalent to the amplitudes \(\sqrt{m_k}\) being the eigenvalues of a \(3\times3\) circulant with off-diagonal-to-diagonal ratio \(1/\sqrt{2}\). We show that the face-centered-cubic (FCC) vacuum lattice of the Selection–Stitch Model supplies this ratio, under a stated ansatz for the family mass operator. Along a \(\langle111\rangle\) axis each site's twelve bonds divide \(6:3:3\) between its own sheet and the sheets above and below, and ABC stacking cycles the three sheet types. If the operator is built from the root-mean-square couplings \(\sqrt{6}:\sqrt{3}:\sqrt{3}\) that a localized state sees, \(K = 2/3\) follows exactly, with no fitted parameter, as a ratio of bond counts. The hierarchy is set by one fitted phase, the gauge-invariant offset of the families' Bloch phase along the stacking axis; the resulting \(m_\tau/m_e = 3477.44\), against \(3477.37 \pm 0.18\) observed, tests Koide's relation rather than the lattice. Among the structures surveyed, only FCC has this property below six dimensions. Delocalized states give \(K = 1/2\), and states in tetrahedral voids, the model's quarks, give \(K \le 0.614\). Realistic lepton mixing requires a neutrino sector that breaks the stacking translation and is complex in the basis of the sheet types.

Study Protocol
Public Health and Healthcare
Public, Environmental and Occupational Health

Aurélie Nakamura

,

Aurore Li

,

Cécilia Samieri

,

Christel Becker

,

Cyrille Delpierre

,

Elodie Faure

,

Elodie Speyer

,

Eloi Chazelas

,

Emeline Lequy-Flahault

,

Emma Juillet

+21 authors

Abstract: The exposome, the totality of environmental exposures encountered throughout life, has emerged as a pivotal framework to complement genomic research and advance our understanding of chronic disease etiology. Building on this paradigm, Inserm launched the BOOSTER Impulsion Exposome program to structure and stimulate national efforts in exposome science through three interconnected work packages (WPs): i) WP1 investigates socio-territorial inequalities in exposures across France by integrating environmental, behavioral, and psychosocial data from national databases and population cohorts; ii) WP2 explores the combined effects of psychosocial and chemical exposures on respiratory health, using an integrative approach that combines epidemiological analyses with experimental and systems toxicology studies in animal models, designing new Adverse Outcome Pathways; iii) WP3 develops advanced causal inference and statistical models to disentangle the complex, multidimensional relationships linking exposome factors to health outcomes. Together, these WPs form a synergistic research ecosystem that bridges social sciences, epidemiology, toxicology, and data science. Beyond methodological innovation, the program contributes to public health by providing tools to identify vulnerable populations, inform prevention strategies, and guide national and European environmental health policies. The BOOSTER initiative thus represents a structural step toward a sustainable, interdisciplinary French exposome research community and a model for future international collaborations.

Review
Environmental and Earth Sciences
Geophysics and Geology

Tanushree Gupta

,

Brijesh Kumar Yadav

,

Maurizio Polemio

Abstract: Groundwater salinization in terrestrial and coastal aquifers is a complex process governed by anthropogenic activities, marine influences, and natural hydrogeochemical processes. Previous studies indicate that freshwater-saltwater interaction (FSI) largely determines the long-term fate of freshwater aquifers. Yet choosing a suitable framework to investigate interactions between subsurface waters of contrasting density remains challenging. This review synthesises studies retrieved from Web of Science, Scopus, ScienceDirect and Google Scholar, using keywords on seawater intrusion, groundwater salinization, hydrochemistry, geophysics, remote sensing, numerical modelling and sandbox experiments. The studies were screened for methodological relevance to FSI and grouped by method and spatial scale. At the regional scale, remote sensing combined with GIS provides only surface proxies (e.g., vegetation and soil salinity indicators) for shallow saline groundwater and requires ground-truthing. At the local scale, geophysical methods are critically assessed for their ability to delineate salinity zones. The review also covers in-situ monitoring, hydrogeochemical analysis, and isotopic tracers, which identify salinity sources and characterise mixing processes. Analytical sharp-interface solutions and index-based vulnerability models serve rapid screening and theoretical benchmarking but are limited to idealised conditions. Numerical tools for predicting FSI behaviour under site-specific conditions are then examined, followed by laboratory-scale physical models, which offer a valuable means of studying seawater intrusion and saline plume migration under changing climate and land use. We compared the strengths, limitations, and combined use of these methods, identify research gaps, and proposed a decision framework that links method selection to the study.

Article
Physical Sciences
Quantum Science and Technology

Kalina Mileva

,

Kliment Naydenov

,

Sia Lozanova

Abstract: This study investigates how integrated terrestrial and satellite quantum key distribution (QKD) architectures may support territorial communication connectivity and regional security in Europe. A structured qualitative comparison considers terrestrial-only, satellite-dependent and hybrid configurations under four disruption scenarios: fibre interruption, persistent optical weather unavailability, optical ground-station failure and trusted-node compromise. Evidence is drawn from peer-reviewed QKD security analyses, terrestrial field networks, satellite-channel studies and space-to-ground demonstrations, with European programme documents used solely for policy and project context. The analysis distinguishes physical connectivity from authenticated, policy-compliant key delivery and identifies six spatial and operational planning criteria: geographic diversity, infrastructure independence, optical availability, trust security, key-management interoperability and tested recovery procedures. Hybridisation offers conditional alternative pathways but cannot itself guarantee continuous service or remove shared trust dependencies. The proposed framework is conceptual; no field measurements, reliability percentages or operational deployment claims are inferred. The contribution is a transparent planning method and an explicit agenda for future quantitative validation.

Hypothesis
Biology and Life Sciences
Life Sciences

Anindya Das

Abstract: The most gruelling scientific question of this century is how life first originated on Earth. In the mid-19th century, Louis Pasteur and John Tyndall disproved spontaneous generation of life through a landmark experiment demonstrating the law of biogenesis: living organisms come only from pre-existing life. However, the concept evolved over centuries, and the general consensus now is that a transition phase existed between abiogenesis and biogenesis, and under radically different environmental conditions, life started from the simplest possible chemical building blocks. Finding a plausible explanation for the emergence of life billions of years ago has produced many interesting hypotheses. However, the basic constituents of life, like carbon, hydrogen, oxygen, nitrogen, phosphorus, and other inorganic substances, combined in a proportion to create the first life on Earth. Simulation studies indicate that abiotic formation of simple organic molecules from carbon dioxide, hydrogen, methane, hydrogen sulphide, ammonia, formate (HCOO-), and cyanide (CN-) is thermodynamically favoured under hydrothermal-vent conditions or in shallow superficial basins and, interacting with catalytic minerals, polymerises into molecules of increasing complexity. To explain the systematic assembling of all the basic structural elements leading to a structural unit with functional autonomy with biochemical synthesis, degradation (metabolism), energy production, self-replication, which can be called “nano-scale organisation”, a hypothesis has been proposed, which is called the “Random Collision, Association, Selection, Rejection and the Acquisition of Added Characteristics Hypothesis”.

Article
Computer Science and Mathematics
Computational Mathematics

Yinze Tang

,

Yihao Li

,

Shengji Wang

,

Pingsheng Zhang

,

Liang Li

,

Zeheng Wang

Abstract: Broadband terahertz (THz) detector design requires linking electromagnetic coupling to electrical response, while repeated full-wave geometry sweeps are computationally demanding. To support design–technology co-optimization (DTCO) of gallium nitride (GaN) THz devices that can significantly reduce the time cost, we present physics-informed electronic design automation (PI-EDA) for GaN high-electron-mobility transistor THz detectors. An expanded electromagnetic-to-electrical dataset supports Lorentz-resonance fitting of 284 full-wave spectra to extract explicit geometry laws, with a learned residual correcting spectral prediction. PI-EDA inverts the resonance law to select sampled geometries, generates a 4 × 5 array plan, and evaluates detectors and readout responses using their full-wave spectra. On held-out width-sweep spectra at 1 V/m incidence, the spectral root-mean-square error is 4.81 V/m, compared with 28.73 V/m for Extra Trees. In a geometry-selection example, automatic arm-length selection places twenty fitted Lorentz centers across 380–1000 GHz with a 1.5 GHz median absolute placement error. A finite-element transistor model and a Dyakonov–Shur solver with experimentally fitted transport parameters evaluate the detector responses for the selected coupler geometries. At equal incident pixel power, the maximum-pixel envelope yields a median pointwise current ratio of 34 over the highest-peak single element. A separate summed-pixel design for 380–430 GHz predicts 6.8 µA at 410 GHz for 100 µW distributed uniformly over the matrix. The operating point is 0.2 V gate overdrive and 20 mV drain bias, with a 10 kΩ transimpedance amplifier predicting a 68-mV output magnitude. PI-EDA links target bands to explicit coupler geometries and predict detector outputs.

Review
Medicine and Pharmacology
Pathology and Pathobiology

Zahra Sedarat

,

Andrew W. Taylor-Robinson

Abstract: Persister cells are a specialized subpopulation of bacteria that enter a metabolically dormant or slow-growing state, rendering them highly tolerant to conventional bactericidal antibiotics. Consequently, they represent a major obstacle to the successful eradication of chronic and recurrent bacterial infections. Persister cells frequently reside within biofilms, thriving inside protective extracellular polymeric matrices. Their inherent phenotypic and physiological heterogeneity further complicates therapeutic targeting. Overcoming the barriers presented by these recalcitrant, non-dividing bacteria requires innovative strategies that go beyond traditional antibiotic design. This review synthesizes current understanding of persister cell biology, including their genetic regulation, toxin–antitoxin systems, the stringent response, and their interplay with biofilms. Furthermore, we evaluate emerging anti-persister cell strategies, classified into resensitization via metabolic stimulation, direct membrane/target destruction, and inhibition of persister cell formation pathways.

Article
Public Health and Healthcare
Public Health and Health Services

Olakunle Daniel Olorunkemi

Abstract: Objectives: To evaluate whether current serum 25-hydroxyvitamin D [25(OH)D] is associated with current hepatitis B surface antibody (anti-HBs) seropositivity among US adults who report hepatitis B vaccination. The response to vaccination was not studied. Methods: This study included 1,382 US adults aged ≥18 years in the 2017–2018 National Health and Nutrition Examination Survey (NHANES) who reported ≥3 doses of hepatitis B vaccine and had measured 25(OH)D and anti-HBs. Survey-weighted logistic regression estimated odds ratios (OR) for anti-HBs positivity by 25(OH)D <50 versus ≥50 nmol/L, adjusted for demographics, income, body mass index, smoking, and season (n = 1,098). For comparison, all 2017–2018 adults regardless of vaccination status were also analyzed. Results: Among vaccinated adults, weighted anti-HBs positivity was 47.8% (95% CI 43.6–52.1), and 30.5% (unweighted) had 25(OH)D <50 nmol/L. Low 25(OH)D was not statistically significantly associated with anti-HBs positivity (adjusted OR 1.12, 95% CI 0.73–1.74), including among anti-HBc-negative adults (OR 1.15, 0.74–1.80). The OR was 1.07 (0.78–1.49) among all 2017–2018 adults. Conclusions: Among adults reporting complete vaccination, current 25(OH)D was not statistically significantly associated with anti-HBs positivity, and results were similar in all adults and in anti-HBc-negative adults. Because 25(OH)D was measured long after vaccination, these data cannot address vitamin D status at vaccination.

Article
Social Sciences
Psychology

Yulu Wang

,

Jianling Gu

Abstract: Current ethical regulations in the field of psychological counseling are predominantly grounded in the principle of “judging by overt conduct”, emphasizing the evaluation and regulation of practitioners’ observable behaviors. However, a persistent gap between ethical knowledge and ethical practice remains prevalent in professional settings: although practitioners are familiar with ethical codes, boundary violations and ethical misconduct continue to occur. Drawing on a publicly discussed ethical controversy in the field of psychological counseling, this paper adopts a psychoanalytic perspective and conducts a dual-level analysis of “overt conduct” and “underlying intent”, proposing an integrated ethical paradigm of “assessing overt conduct and underlying intent.” Ethical practice in counseling and psychotherapy requires both behavioral accountability and sustained awareness of counselors' unconscious psychological dynamics, such as personal issues and desires. Furthermore, a dual-dimensional Awareness-Desire Matrix is proposed to identify counselors’ psychological positions in ethical practice. This framework aims to bridge the gap between ethical compliance and ethical internalization, providing practical implications for counselor training, supervision systems, and professional regulation.

Article
Social Sciences
Education

M. A. Muazar Habibi

,

Ika Rachmayani

,

Abdul Kadir Jaelani

,

Baiq Nada Buahana

Abstract: Child-centred disaster risk reduction (DRR) has reframed children as learners capable of protective action, yet hazard and risk assessments rarely reach early childhood classrooms in multi-hazard regions of the Global South. This study examined how risk information was translated into an eight-week, culturally grounded, play-based DRR curriculum for 60 children aged 5–6 years on Lombok, Indonesia, an island struck by a destructive earthquake sequence in 2018. A mixed-methods curriculum-development design combined risk documents, aggregate pre–post observation indicators, interviews with six teachers and three Sasak community leaders, and curriculum records. Stories, songs, role play, calm simulations, and family tasks incorporated the Sasak values of awik-awik (shared customary agreements) and gotong royong (mutual cooperation). Reported indicators rose from 45% to 80% for risk understanding, 40% to 81% for evacuation preparedness, and 50% to 85% for simulation response. Five themes concerned safety as a shared agreement, learning through the body, familiar places as learning anchors, children as active rehearsers, and the home as a second classroom. These informed a six-domain Early Childhood DRR Curriculum Framework for West Nusa Tenggara (ECDRRCF-NTB). Aggregate data and the absence of a comparison group preclude causal claims; the framework is a curriculum proposition for controlled evaluation.

Article
Social Sciences
Urban Studies and Planning

Pema Thinley

,

Christine Steinmetz-Weiss

Abstract: Presented as case study research, the study applied a performance measurement framework to assess the implementation of Thimphu City’s master plan, the Thimphu Structure Plan 2023. The study contributes to urban planning practice and academic discussion on the limited subject of goal-oriented performance evaluation of cities’ master plans. Through a systematic literature review of methods aligned with the regenerative and distributive principles informing TSP 2023, the Urban Planning Implementation Evaluation Framework was identified as the most suitable method because it can assess both conformance and performance. Its application revealed moderate conformance and low performance, requiring urgent intervention from the city administration. This study provides Bhutanese government agencies with the first adaptable, mathematical tool for coordinated, transparent, and evidence-based periodic assessment of TSP 2023 objectives, with potential application to master plans beyond Bhutan.

Article
Computer Science and Mathematics
Algebra and Number Theory

Eldar Sultanow

Abstract: We study why Collatz-type maps with similar local predecessor statistics can have different global cycles and basin shares. The maps 3\( x \) + 1 and 3\( x \) - 1 have corresponding local branching and cotree statistics, yet the latter has three known positive cycles. The classical parity-word divisibility criterion gives an exhaustive census of 1097 primitive cycles for 3\( x \) + \( c \) with \( |c|\le1000 \) and parity-word length at most 40. Using first-passage estimates of Gonçalves, Greenfeld and Madrid, we prove that every completely invariant set of the accelerated map \( T_{3,c} \), for odd \( c\ge-1 \), has a logarithmic density. In particular, basin and predecessor-set densities exist in this sense. Conditionally on a recent natural-density transport of Allikvere, which carries over to all these maps, they are natural densities. The resulting predecessor potential \( h(a) \) satisfies an exact conservation law across the backward levels of a cycle. Gcd classes and multiplicative cosets give further invariant sets of exactly computable density, but do not by themselves determine individual basin densities. We also prove a residue-class lower bound for the means of \( a\,h(a) \) in terms of the Syracuse law; a matching upper bound and the proposed exact profile remain conjectural. An appendix reduces part of the missing upper bound, conditionally on the coupling estimates of that proof, to uniform integrability of the stopped predecessor potential of the odd targets; large-scale predecessor counts are consistent with this but cannot decide it. Numerically, a parameter-free harmonic backward-mass model predicts 410 measured basin shares with approximately 3% median relative error. Additional conclusions for \( 3x\pm1 \) use recent predecessor-density preprints of Mazur and El Jaouhari. Lean 4 checks the elementary finite statements; the density and Syracuse statements are not machine-certified.

Review
Arts and Humanities
Museology

Markos Konstantakis

Abstract: Collections of galleries, libraries, archives and museums (GLAM) are increasingly crawled, ingested and restated by external AI systems. We conducted a registered rapid scoping review (JBI methodology, PRISMA-ScR; OSF 10.17605/OSF.IO/WEJ4K) to map evidence on the relationship between GLAM collections and the AI systems that consume or reproduce them across readiness, access, fidelity, regulation and equity. OpenAlex, Crossref and arXiv were searched for 2019–2026, supplemented by snowballing and grey literature. Screening and charting were AI-assisted, with separate blinded AI passes to check agreement. From 6763 database records screened, together with snowballing and grey literature, 411 sources were included; 82% were published in 2025–2026. The evidence is overwhelmingly inward: 343 sources concern AI applied by institutions to their own collections, and only 23 empirical sources address external AI systems. Grounding in collection data mostly improves the accuracy of institutional systems, but the accuracy of general-purpose AI about collections rests on a few benchmarks, no eligible study of what external AI says about institutions was identified, crawler load is documented through self-reports, and legal duties, access models and community consent are analysed but rarely measured. We propose an eight-item research agenda with evidence labels and indicators, and a readiness checklist for institutions.

Review
Business, Economics and Management
Economics

Yunzhang Li

,

Xinlei Pan

,

Tao Ding

Abstract: Blue carbon is increasingly presented as an investable climate solution, yet large coastal carbon stocks do not necessarily generate additional mitigation, saleable credits, or financially viable restoration. We develop a four-account framework that separates biophysical carbon, creditable mitigation, project cash flow, and social welfare, and apply it to the conversion of ecological value into credible and durable economic benefits. The synthesis connects carbon accounting, ecosystem-service valuation, trading incentives, restoration finance, and coastal governance. Four analytically distinct accounts are proposed: biophysical carbon, creditable mitigation, project cash flow, and social welfare. Their separation explains why high ecosystem values can coexist with weak commercial returns, and why low-cost credit supply is not sufficient evidence of environmental integrity. Comparative evidence indicates that project viability depends on counterfactual land use, sediment-carbon uncertainty, restoration trajectories, transaction costs, contractual demand, tenure, and the distribution of benefits. Game-theoretic studies identify plausible incentive mechanisms, but their equilibria should not be interpreted as observed market performance. Climate change and coastal development further affect the quantity, timing, and durability of prospective revenue. An integrated decision framework therefore prioritizes credible accounting and rights before financial structuring, matches financing instruments to project maturity, and evaluates portfolios against correlated ecological and institutional risks. The central policy implication is that carbon markets should complement, rather than substitute for, public investment in coastal protection, biodiversity, and community stewardship.

Article
Biology and Life Sciences
Biochemistry and Molecular Biology

Fabiana Uomo

,

Chiara De Falco

,

Ferdinando Barretta

,

Nancy Messina

,

Rosa Redenta De Simone

,

Daniela Dottore Stagna

,

Antonia Pascarella

,

Enza Mozzillo

,

Dario Iafusco

,

Giulia Frisso

+1 authors

Abstract: Background/Objectives: Mitochondrial diseases (MDs) are heterogeneous genetic disorders caused by oxidative phosphorylation impairment, governed by both mitochondrial DNA (mtDNA) and nuclear DNA (nDNA). Traditional multi-step diagnostic strategies, involving mtDNA long-range PCR and screening of a limited set of nuclear genes, often leave many patients without a molecular diagnosis. This study evaluates the clinical implementation of a single-step MITO-NUCLEAR Next-Generation (NGS) panel (whole mtDNA and 254 nuclear genes), previously developed and validated by our group. Methods: Buccal swab DNA from 70 patients with suspected MDs was sequenced using an optimized probe dilution ratio (1:500, mtDNA: nDNA), achieving mean depths of ~10,000× (mtDNA) and 110× (nDNA). Variants were classified according to ACMG/AMP guidelines and their mtDNA-specific specifications. Results: Ninety unique variants were identified (86.7% nuclear). A definitive molecular diagnosis was reached in 17% of patients, compared with 5.7% for mtDNA analysis alone: four primary mtDNA defects (MT-TS1, MT-CYB, MT-TL1, MT-ATP6), two nuclear-encoded mitochondrial disorders (ECHS1, FH) and six non-mitochondrial phenocopies (WFS1, ATP7B, G6PC1, ABCA4, KIF1A). A possible diagnosis was reached in 20% of patients, 46% were carriers of monoallelic variants in recessive genes, and 17% were negative. Conclusions: Simultaneous MITO-NUCLEAR NGS testing significantly increases diagnostic yield over single-genome screening. This single-step targeted panel offers a fast and efficient tool that facilitates variant identification across both genomes, supporting accurate clinical management and timely familial genetic counseling.

Article
Engineering
Civil Engineering

Malek Alazzam

,

Khawlah Harahsheh

Abstract: The increasing adoption of artificial intelligence in construction management has created opportunities for data-driven schedule-risk prediction. However, the reliability and generalizability of machine-learning models across heterogeneous infrastructure projects remain insufficiently understood, particularly when project-level variability and temporal changes are not adequately considered. This study investigates the predictive performance, probability calibration, and transferability of machine-learning models for identifying activity-duration growth in infrastructure construction projects. Using an extensive dataset comprising 148,087 construction activities across 271 infrastructure projects, the study evaluates logistic regression and gradient-boosting models under conventional and project-balanced learning strategies. A rigorous evaluation framework incorporates five-fold project-held-out validation, project-cluster bootstrap uncertainty estimation, probability calibration, temporal validation, explainable artificial intelligence, and risk-based activity prioritization. Results indicate that standard gradient boosting achieves an area under the receiver operating characteristic curve (AUROC) of 0.633, while project-balanced learning achieves comparable discrimination (AUROC = 0.632) and improved project-level probability accuracy. However, temporal validation reveals substantial performance deterioration for certain model configurations, highlighting the influence of distributional shifts and project heterogeneity. Furthermore, model effectiveness in prioritizing high-risk construction activities varies across project sizes. These findings demonstrate the importance of project-aware evaluation, temporal robustness assessment, and uncertainty quantification when developing intelligent schedule-risk prediction frameworks. The study contributes methodological and practical insights for evaluating data-driven construction risk management systems and supporting more reliable infrastructure project decision-making.

Article
Computer Science and Mathematics
Robotics

Jungmo Kang

Abstract: Open-source LiDAR localization tools are organized by sensor suite. We describe PRISM-Loc, an open-source ROS 2 stack that puts three established estimator families behind one contract: laser2d, a likelihood-field Monte Carlo localizer; ndt3d, a planar NDT-MCL backend (Saarinen et al., 2013); and fusion3d, a 15-state error-state Kalman filter (ESKF) fusing LiDAR, IMU, and RTK-GNSS. All backends publish the same map→odom transform and pose with covariance and accept the same /initialpose. The filter mathematics live in Eigen-only C++17 cores tested without ROS, and laser2d relocalizes by branch-and-bound scan matching with multi-scan verification that can report a place as ambiguous. We claim no new estimator theory. Seeded synthetic experiments show a 0.105 m mean tracking error, 97 of 100 room relocalizations, verification cutting false accepts on repeated-structure maps from 21 to 1 of 240 queries, and an overconfident ESKF covariance. On public data we test only starts from a wrong initial pose. On NCLT, ndt3d with the ground grid and tempered likelihood converges from wrong initial headings where a 6-DoF scan-to-map NDT tracker does not (2 m and 45◦: 12 of 12 runs against 4 of 12; 2 m and 90◦: 12 against 0; 10 m and 90◦: 11 against 0), although the particle filter is given the initial spread and the converged tracker is more accurate (median RMSE 0.15 against 0.30 m). On the MIT Stata Center dataset, branch-and-bound relocalization finds the pose in 9 of 12 runs against 3 of 12 for nav2_amcl global localization, and laser2d converges faster than AMCL in the runs where it converges, but AMCL converges as often or more often from offset starts. AMCL with Nav2 defaults fails on the sparse NCLT 2D projection even without offset. fusion3d has no public evaluation. Samples are small, Stata odometry is synthesized, and no field-accuracy claim is made. Code: https://github.com/kjungmo/prism_loc.

of 6,700