Search Results (136)

Background: Adult hearing loss is common, yet hearing-aid uptake remains inadequate due to systemic and financial barriers in traditional audiologist-based fitting models. Over-the-counter (OTC) and self-fitting strategies have emerged to improve accessibility, allowing users to select and adjust their devices independently. However, concerns remain that reducing professional involvement may compromise electroacoustic accuracy, speech-in-noise outcomes, and overall patient satisfaction. Methods: A comprehensive literature search across PubMed, Web of Science, CENTRAL, Scopus, and Embase was conducted up to 1 June 2026. Parallel-group and crossover randomized controlled trials (RCTs) comparing self-fitting with professional-fitting were included. The primary outcome was patient-reported hearing benefit. Standardized mean differences (SMDs) and risk ratios (RRs) were pooled using a random-effects model. Results: Seven RCTs involving 1092 participants were included. The analysis revealed no statistically significant difference between self-fitting and professional-fitting strategies regarding patient-reported hearing benefit (SMD: 0.01, 95% CI [−0.44, 0.46], p = 0.97, I² = 89.2%, very low certainty of evidence). Similarly, no significant differences were observed across secondary outcomes, including speech-in-noise performance (SMD: 0.11, 95% CI [−0.25, 0.47]; p = 0.55), hearing loss (p = 0.12), hearing-aid satisfaction (p = 0.31), device adherence (p = 0.65), or the need for follow-up adjustment (RR: 0.46, 95% CI [0.12, 1.71]; p = 0.25). Safety outcomes were sparse and did not demonstrate a statistically significant difference between the fitting approaches; however, the evidence was imprecise and inconclusive. Conclusions: Self-fitting hearing aids showed no clear differences from professional best-practice fitting across several patient-reported and functional outcomes in adults with uncomplicated mild-to-moderate hearing loss. However, the certainty of evidence was very low because of risk of bias, substantial heterogeneity, imprecision, and the small number of trials. These findings are therefore hypothesis-generating and suggest that well-designed self-fitting pathways may have potential as an accessible alternative for selected users, but larger, rigorous, and longer-term trials are needed before broader clinical implementation can be recommended. Full article

Financial loss, data theft, and user harm can result from phishing attacks, which are increasing in frequency and variety beyond traditional email-based attacks to include mobile forms like smishing, quishing, malicious mobile applications, and mobile phishing websites. While a number of reviews have already addressed phishing, Android Mal-Ware, and Explainable Artificial Intelligence (XAI), the literature is still disjointed in the lightweight deployment, model explainability, temporal attack behaviour, coordinated phishing detection, and evaluation datasets of intelligent mobile systems. The purpose of this systematic literature review (SLR) is to analyse all the major phishing detection methods using deep learning published from 2020 to 2026 and to focus on the key features of these methods, such as the model architectures, the mobile deployment, the explainability, the temporal modelling, coordinated detection, and the evaluation datasets. Studies were identified, screened, and assessed for eligibility using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines before being included in the review for qualitative synthesis. Out of 101 studies that were included in the final review, the evidence was comparatively synthesised considering the following: learning architecture, detection modality, deployment characteristics, explainability mechanisms, the ability to model time, and the evaluation performance reported. The synthesis results show that phishing detection is generally good, with the accuracy of the transformer-based, hybrid, and attention-enhanced architectures being above 97% for some of the selected studies, and lightweight and edge-oriented architectures making the deployment more feasible for resource-constrained mobile environments. The reviewed evidence also demonstrates the increasing combination of XAI, temporal sequence modelling and multimodal analysis to enhance the transparency and robustness of detection. Comparisons between studies are, however, restricted by variations in the datasets, the representation of the features, the evaluation protocols, and the performance metrics. The results highlight the importance of future research focused on explaining lightweight, temporally adaptive, and robust phishing detection frameworks for intelligent mobile systems. Full article

The Kerper–Bowron (KB) Method (patent pending) projects expected claims at the individual contract level. This paper extends that cash-flow engine to manufacturer-warranty accruals and to three proposed uses: collateralized risk transfer, lending against service-contract equity, and risk-adjusted customer lifetime value (CLV). Manufacturer warranties and separately priced service contracts are treated as related but distinct products under ASC 460, ASC 450, and IAS 37. Expected cost per unit of exposure is formed with a generalized linear model; a Tweedie mean–variance function is used as a working choice, not as a tested warranty distribution. Accident-month estimates are allocated to payment months, incurred-but-not-reported cost on pre-valuation months is isolated, and remaining paid cash flow is split into pre-valuation runoff and post-valuation occurrence. One present-value risk margin is taken from the predictive distribution as the present value of the gap between a stated percentile and the mean; a constant loading on the discounted mean is an illustrative substitute when simulation is not run. The contribution is contract-level granularity and a single paid path that can be refreshed as experience and assumptions change. The same paid path can be used as a financial-monitoring tool: expected claims, equity, and risk-adjusted values can be refreshed as time passes, actual results emerge, and model or economic assumptions change. Accuracy, balance sheet derecognition, investor diversification, and lendable capacity would be the subject of further research. Full article

How much a college education pays off varies widely from one student to the next, and that variation matters for admissions, financial aid, and mobility policy. Most estimates, however, report a single average return and treat each institution as an isolated row in a table, ignoring how schools relate to one another and how outcomes are distributed across demographic groups. We present FairEdu-GCT (Graph-Enhanced Causal Transformer), a framework that couples a heterogeneous graph encoder with a Transformer sequence model and an explicit fairness penalty. Institutions, geographic regions, and academic disciplines form a typed graph whose edges record graduate flows, spatial proximity, and disciplinary overlap. A Relational Graph Attention Network (R-GAT) turns this structure into institutional ecosystem embeddings that carry peer effects, regional labor-market signals, and the spread of institutional prestige, none of which survives in tabular representations. A Transformer then encodes each student’s educational history and merges it with the institutional embedding through a cross-modal attention bridge, and a counterfactual decoding head returns the full conditional earnings distribution under alternative institutional choices rather than a single point estimate. Because students are not randomly assigned to schools, we make no claim of strict causal identification; we treat CATE and PEHE strictly as estimation-quality diagnostics for a confounding-adjusted contrast, not as evidence of a proven causal effect. We instead adjust for observed confounders through a doubly robust objective and add an equalized opportunity regularizer so that accuracy does not come at the expense of protected subgroups. On linked U.S. College Scorecard, IPEDS, and NLSY97 data (6256 institutions drawn from 7312 Title IV schools; 161,043 person-institution-year records from 8984 respondents followed for ten years), FairEdu-GCT lowers RMSE by 16.2% and Precision in Estimation of Heterogeneous Effects (PEHE) by 25.1% against the strongest baseline (Causal Forest, DragonNet, TARNet, and TabTransformer), and shrinks the demographic parity gap by 46.5%. All gains are reported with 95% confidence intervals and effect sizes over ten seeds, and an extended fairness audit (calibration, predictive parity, and subgroup robustness) confirms the improvement is not confined to the two metrics we optimize. Ablations attribute a 9.0% RMSE reduction to the R-GAT encoder alone. A group-conditional SHAP analysis further shows that graph-derived proximity to regional technology clusters is an unusually strong predictor for first-generation minority students, a signal that tabular features cannot recover and one we read as a within-model association rather than a policy lever. Full article

Whether machine-learning models extract predictive signal from ETF returns across markets at different efficiency levels and whether statistical accuracy translates into trading value, remain contested. We examine this for iShares MSCI Brazil (EWZ) and iShares Core S&P 500 (IVV) from January 2010 to July 2026, training through December 2022 and testing thereafter. Random Forest, XGBoost with random search, XGBoost with Bayesian optimization, LSTM, GRU and an LSTM + XGBoost ensemble, are compared against historical mean, random walk, and AR(1) benchmarks at one-day (h = 1), five-day (h = 5) and monthly (h = 21) horizons using ten technical indicators. Every model is also evaluated against the classifier that predicts the majority class, and risk-adjusted performance is reported with bootstrap intervals. No model exceeds that trivial classifier in any combination examined. The two markets fail by distinct mechanisms: collapse onto the majority class in the developed market, and dispersed but unprofitable signals in the emerging one. Under Diebold–Mariano tests with autocorrelation-consistent variance and false-discovery control, no model is superior to the historical mean. No strategy outperforms Buy-and-Hold, and in the emerging market, no Sharpe ratio is distinguishable from zero. Where directional significance does appear, at the monthly horizon in the emerging market, it delivers no economic value. Together, these results argue for evaluating financial forecasting models simultaneously on regression metrics, economic performance, and regime stability rather than on any single criterion. Full article

Financial question answering is often treated as document question answering, although financial evidence is both multimodal and time-dependent. Semantically equivalent facts expressed in narrative text, tables, page images, or Extensible Business Reporting Language (XBRL) should support consistent answers, whereas a disclosure may support a query only after becoming public. We formalise this combination as crossmodal evidence symmetry under a causal temporal boundary and introduce TempFinRAG, a multimodal temporal retrieval-augmented generation (RAG) framework for point-in-time financial question answering. Given a question, company, and as-of date, the framework enforces the information boundary defined by U.S. Securities and Exchange Commission (SEC) filing availability; aligns page layout, text, table structure, and XBRL facts; retrieves time-valid evidence; executes auditable financial calculations; and generates a cited answer. A verifier checks temporal validity, claim support, numerical consistency, and the need to abstain. We further introduce TempFinQA, a point-in-time evaluation protocol built from public filings and XBRL facts, and evaluate the framework on complementary evidence-grounded, numerical, conversational, and multi-table benchmarks. On TempFinQA, TempFinRAG improves answer accuracy from 66.7% to 78.9% over hybrid RAG while reducing temporal evidence leakage from 10.8% to 1.7% and hallucination from 17.3% to 7.9%. Reliable financial question answering therefore requires consistent treatment across evidence representations and deliberately asymmetric access across time. Full article

Voice phishing (vishing) causes financial and psychological harm, yet deployed detectors that scan transcripts for scam vocabulary are more fragile than their benchmark scores suggest: their near-perfect accuracy on standard Korean corpora is memorization, collapsing under scammer paraphrase or language-model rewriting. We construct KorMMP, a benchmark keeping real regulator-sourced scam audio and real benign speech but decorrelating transcript from label, and MiLTL, an on-device detector built on affective and neutrosophic channels, including a cross-modal contradiction signal (XM) formulated to weigh lexical warmth against vocal coldness. In the reported evaluation, XM’s measured contribution is confidence banding and escalation routing rather than ranking accuracy. A scoring rule with zero gradient-learned parameters at inference screens every segment, referring ambiguous calls to a small on-device language model; raw audio stays on the device. We measure both stages on a commodity CPU container and smartphone; these budgets cover the detector only and exclude speech recognition, which we expect to dominate a live end-to-end budget. With no training on the hard benchmark, MiLTL achieves an AUROC of 0.965 (recording-level cluster bootstrap 95% CI 0.943–0.984), while every evaluated comparator stays below 0.69: text-only, audio-only, fusion, 7B multimodal. MiLTL scores lower on the saturated corpus, as expected for a detector designed not to rely primarily on lexical shortcuts. KorMMP is a controlled stress test of lexical decorrelation, not a measure of in-the-wild detection; its harmful and benign audio come from different corpora, so the source and label are structurally confounded, and the residual source effects cannot be fully excluded. XM is author-defined, constructed rather than naturally occurring in the synthetic stratum and not yet validated against human perception; this is the principal open limitation of the work. Within that scope, the work provides a reproducible basis for on-device vishing defense and a benchmark for detector robustness under vocabulary shift. Full article

Financial media risk is jointly driven by multisource content, including news reports, corporate announcements, social media posts, short videos, and livestreams, while content authenticity, propagation velocity, asset relevance, and trading infrastructure conditions can simultaneously influence short-term market fluctuations. To address the limitations of existing methods, including their reliance on either textual information or market sequences, insufficient source verification, and inadequate alignment of asynchronous multimodal signals, FMRP-Net is proposed for artificial intelligence-driven sensing. Event semantics, risk categories, and asset association information are first extracted through a pretrained financial language model and cross-modal consistency analysis. A dual-layer hardware reliability perception module is then employed to integrate sensing evidence from cameras, microphones, terminal inertial signals, server temperature, power consumption, network traffic, and transmission latency. Heterogeneous temporal propagation graphs, cross-frequency alignment, and bidirectional propagation–market coupling are further incorporated to jointly predict market direction, volatility, risk level, and propagation trends. Experimental results demonstrate that FMRP-Net achieved an Accuracy of $0.832$, a Macro-F1 of $0.824$, a ROC-AUC of $0.891$, an MCC of $0.665$, and a PR-AUC of $0.883$ for market direction prediction over future horizons of 5, 15, 30, and 60 min, indicating a balanced performance in terms of Precision and Recall. For volatility prediction, MAE, RMSE, and MAPE values of $0.0178$, $0.0271$, and $12.46\%$ were obtained, respectively, together with an $R^2$ of $0.812$. In the ablation study, the media risk Macro-F1 and source reliability AUC reached $0.842$ and $0.929$, respectively, while the propagation-scale prediction error was reduced to $0.109$ and the average early-warning lead time reached $10.6$ min. These results demonstrate that the integration of multimedia semantics, hardware sensing evidence, and propagation structures can effectively improve the accuracy, stability, and interpretability of financial market prediction and risk early warning. Full article

The integration of artificial intelligence (AI) and data science into organizational evaluation practices creates socio-technical systems in which human rating behavior, organizational templates, regulatory pressure, and analytical algorithms jointly determine what can be measured and learned. This single-case study examines the measurement of Environmental, Social, and Governance (ESG) compliance in financial software engineering projects, treating one firm’s project evaluation practice as a socio-technical system and applying an open-source data science and machine learning workflow to 207 anonymized archival project records. Correlation analysis revealed near-unity associations between the stakeholder-rated social and governance factors and the overall project score (r = +0.995 and +0.966, p < 0.001), while the environmental factor was unrelated to the score; post-hoc diagnostics (a seven-component principal-component structure, Harman’s screen, selective near-zero same-source correlations, and marker-variable partial correlations) bind, but cannot eliminate, method-based explanations, so the coefficients are interpreted as a descriptive property of the firm’s evaluation system rather than as estimates of relationships between validated, distinct constructs. Exploratory machine learning classifiers performed weakly—kNN at chance (AUC = 0.497) and SVM only modestly above the no-information baseline (accuracy 61.8%)—a result consistent with the constraints that the social subsystem imposes on the learnability of the records it generates, although technical factors, including the dichotomization of the target variable, the modest sample size, and model configuration, cannot be ruled out as contributing explanations; descriptive statistics are reported for all variables, and a diagnostic analysis reconciles the apparent divergence between the near-unity correlations and the weak classification performance by showing that the two rest on different feature sets, the near-redundant social and governance ratings having been withheld from the classifiers. The findings offer a proof of concept and a structured agenda for AI-enabled, project-level ESG measurement in socio-technical systems. Full article

In incentive-based demand response, an aggregator is paid for the gap between a customer’s metered load and the baseline load that would have occurred without a curtailment signal. This baseline is never recorded during the event, yet the settlement depends on it, so it must be reconstructed from the load observed before and after the curtailment window. At the portfolio level on which settlement is cleared, this amounts to filling a single contiguous gap, aligned with the daily peak, in an otherwise complete record. To estimate the portfolio-level customer baseline load (CBL), we propose AFCANet, which folds the one-dimensional CBL time series into a period-aligned two-dimensional tensor whose two axes describe different things. The intra-period axis traces the shape of a single daily cycle, which is locally smooth and strongly autocorrelated, while the inter-period axis links the same clock time across successive days, a longer-range and less locally smooth dependency. At the core of AFCANet is the Axis-Factorized Convolution–Attention (AFCA) block, which assigns a convolution to the intra-period axis, where its locality and weight-sharing suit the smooth daily shape, and self-attention to the inter-period axis, where its ability to link distant positions suits the cross-day dependency. Experiments use metered load from the Low Carbon London trial dataset with half-hourly resolution, evaluated under a control-group protocol in which the masked baseline is directly verifiable; AFCANet attains a MAE of 20.88 kWh, a MAPE of 1.50%, and a near-zero bias of 4.56 kWh, improving on averaging, regression, and learning-based imputation baselines. A controlled ablation that swaps the two operators confirms that the matched axis assignment is the source of the gain. Since the evaluation relies on synthetic curtailment windows in which no behavioral response is present, the reported accuracy should be read as an upper bound on the performance attainable in live demand-response events. The near-zero bias is of direct practical value to load aggregators, as a baseline free of systematic over- or under-estimation supports accurate curtailment measurement and fair financial settlement in incentive-based demand response. Full article

Housing market research has traditionally emphasized forecasting continuous price levels, often overlooking the discrete regime shifts that more directly capture cyclical risk and market turning points relevant to construction planning and investment decisions. This study develops an interpretable machine learning framework to forecast monthly U.S. housing market expansion and deceleration regimes one month ahead by integrating macroeconomic, financial, and construction-related indicators with national housing price data spanning January 1993 through March 2025. Feature selection and hyperparameter tuning are conducted entirely within the training sample using time-series cross-validation, ensuring that all reported performance metrics reflect genuine out-of-sample generalization. Recursive feature elimination combined with variance inflation factor screening yields a compact, seven-variable predictor set, with no macro-financial variable contributing an incremental discriminatory signal. Five classification models spanning linear and tree-based ensemble families are benchmarked under a strict temporal 80%:20% train–test split. Tree-based ensemble models consistently outperform the linear baseline, with Random Forest achieving the highest holdout AUC (0.932) and the most consistent deceleration detection across nested cross-validation folds. To ensure methodological transparency, explainable AI techniques, including SHapley Additive exPlanations, partial dependence, and individual conditional expectation analyses, are employed to interpret both global and local predictive associations underlying regime classification. Construction-related indicators, particularly Construction Put in Place and Building Permits at short lag horizons, emerge as the dominant supply-side predictive signals, outperforming macro-financial variables in regime discrimination by a margin of 0.184 in training CV AUC. By shifting the analytical focus from price forecasting to one-month-ahead regime prediction and integrating predictive accuracy with economic interpretability, this study provides a transparent and scalable framework for monitoring housing market cycles with direct applications to construction risk management, procurement timing, and project planning. Full article

Credit risk assessment is a core component of financial decision-making. This study develops an explainable machine learning framework for modeling loan approval decisions on heterogeneous tabular data, centered on a Cross-Attentional Tabular Transformer that applies bidirectional cross-attention between numerical and categorical feature groups. The prediction target is historical loan-approval status, treated as a proxy for, not a direct measure of, borrower default risk; a supplementary validation on a dataset with an authentic default label is also reported. Class imbalance is addressed through focal loss, and post hoc interpretability is provided through SHAP analysis. Three classifiers, Random Forest, Gradient Boosting, and the proposed transformer, are evaluated on a 5000-sample credit dataset using accuracy, precision, recall, F1-score, ROC-AUC, and average precision. Gradient Boosting achieves the best performance (accuracy 0.9640, F1-score 0.9189), with Random Forest comparable; the proposed transformer reaches 0.9530 accuracy and 0.8949 F1, without surpassing the ensembles and at substantially higher computational cost. A five-split robustness comparison additionally evaluates XGBoost, LightGBM, CatBoost, and calibrated logistic regression: all three Gradient-Boosting variants and both classical ensembles exceed the transformer’s performance on every metric, while calibrated logistic regression does not. The evaluated baseline set excludes deep tabular architectures such as TabNet, FT-Transformer, SAINT, and TabPFN-style methods. Across the three primary classifiers, SHAP identifies credit score, employment status, and income as the dominant features, consistent with domain expectations. The results characterize the observed performance–efficiency trade-off between ensemble methods and attention-based tabular learning under the evaluated data conditions. Full article

Claim reserving is one of the most important tasks in non-life insurance, because it directly affects solvency assessment, financial reporting, and risk management. Traditional reserving methods typically rely on assumptions of relatively homogeneous claim-development processes and may fail to capture hidden structures within complex insurance portfolios. This paper presents a reserving framework that integrates Topological Data Analysis (TDA) with both aggregate and micro-level reserving methodologies. Using a portfolio of motor insurance claim payments as an empirical case study, we employ topological techniques to identify latent claim-development regimes and portfolio heterogeneity. The extracted topological information is subsequently incorporated into a TDA-enhanced Chain-Ladder (CL) methodology and an Inverse Probability Weighting (IPW) reserving framework. The empirical results indicate that the proposed TDA-based approaches improve reserve estimation accuracy relative to their traditional counterparts. Both TDA-CL and TDA-IPW produce reserve estimates that are closer to the realized future claim payments than those obtained using their traditional counterparts. The findings suggest that topological structures contain valuable information for actuarial reserving and that Topological Data Analysis may offer a useful new perspective for actuarial reserving research. Full article

This study provides an exploratory predictive analysis to examine how different dimensions of digital infrastructure—capital market development, digital payment adoption, e-commerce penetration, and market volatility—predict the financial performance metrics of fintech ventures in Egypt. Using panel data from ten fintech ventures listed on the Egyptian Stock Exchange over the period 2017–2023, the research employs Random Forest machine learning algorithms alongside Logistic Regression as a baseline comparator. Feature importance analysis identifies the most significant predictors of profitability across four performance metrics: gross revenue, sales growth, gross margin, and net profit margin. This study employs Random Forest with five-fold cross-validation. Hyperparameters were optimized via grid search, and feature importance scores are reported with cross-validation standard deviations. To address panel structure concerns, we additionally employ leave-one-firm-out cross-validation. All findings reflect predictive associations only; no causal claims are made due to potential reverse causality. Findings show that capital market development emerges as the most important predictor across all profitability metrics, accounting for 45% of feature importance for net profit margin and 42% for gross revenue (mean importance across five folds; SD = 0.07–0.08). Digital payment adoption exhibits a paradoxical dual association—positively associated with revenue and margins through operational efficiency (38% importance for gross margin; SD = 0.08) while negatively associated with sales growth (22% importance; SD = 0.10). Gross online sales show limited predictive efficacy, affecting only gross margin. Market volatility correlates solely with sales growth. Random Forest consistently outperforms Logistic Regression across all models, with accuracy rates ranging from 68% to 76% (compared to a chance level of 50% and a majority-class baseline of 52–58%). Due to the limited sample of 70 firm-year observations, these findings must be interpreted as strictly exploratory and hypothesis-generating; they apply uniquely to publicly listed fintech firms on the Egyptian Stock Exchange and cannot be generalized to private, early-stage, or unlisted fintech startups without further empirical validation. Full article

Employee attrition is a significant organizational challenge associated with substantial financial costs and the erosion of institutional knowledge. This study presents an AI-based Management Information System (MIS) that integrates machine learning (ML) models to forecast employee turnover and support technical interpretability for HR decision-making. Using the IBM HR Analytics Dataset comprising 1480 employee records with 38 features, we implemented a rigorous preprocessing pipeline—including Synthetic Minority Over-sampling Technique (SMOTE) applied exclusively within training folds to prevent data leakage, one-hot encoding, Z-score normalization, and mean-value imputation. Four ML classifiers—Logistic Regression (LR), Random Forest (RF), Multi-Layer Perceptron (MLP), and XGBoost—were evaluated under a stratified 80/20 split with 5-fold cross-validation. XGBoost achieved the highest performance, attaining an accuracy of 87.83%, a ROC-AUC of 0.94, a PR-AUC of 0.96, and an F1-score of 93.04%, attributed to its sequential boosting mechanism and built-in L1/L2 regularization. Beyond predictive performance, the system incorporates SHapley Additive exPlanations (SHAP) to deliver feature-level transparency, enabling HR professionals to engage in proactive, informed retention interventions while retaining full decision-making authority. Within-dataset comparisons confirm that the proposed framework outperforms prior methods evaluated on the same benchmark; cross-study accuracy comparisons are reported as contextual reference only, given differences in datasets and experimental protocols. The system facilitates human oversight by positioning AI as a decision-support collaborator rather than an autonomous replacement in workforce management. Future work will address real-time deployment, controlled user studies with HR practitioners, and validation with actual organizational HR data. Full article