Search Results (14,941)

Tax administrations increasingly use data-driven risk models to prioritize collection resources, yet machine learning applications to personal income tax (PIT) arrears remain limited. This study introduces TaxMind (v1.0), an interpretable risk management framework for predicting whether PIT obligations will progress to mandatory collection and for integrating debtor- and debt-related information into risk-based segmentation. Administrative records from the Lebanese Tax Administration were analyzed; after removing 5385 exact duplicates from 16,010 records, the final dataset contained 10,625 obligations across 9264 taxpayers. Five tuned classifiers were evaluated using a taxpayer-grouped train/test design, with SHAP used for model interpretation. XGBoost achieved the highest observed discrimination (ROC-AUC = 0.784; accuracy = 0.708; F1-score = 0.704), closely followed by Random Forest (ROC-AUC = 0.780); the leading tree-based models substantially outperformed logistic regression benchmarks. SHAP identified Total Tax Amount, Tax Category 2, and Age as the three leading individual encoded features, while the Age contribution was nonlinear and varied across Tax Categories. The principal SHAP ranking was highly consistent under Random Forest. Top-decile ranking by predicted probability captured only 0.4% of the monetary exposure of realized mandatory collection cases, versus 92.7% under probability-weighted exposure ranking. These findings show that debt-related information remains central, but selected debtor characteristics add predictive value, supporting TaxMind as a model-agnostic early-warning framework for preventive tax debt management. Full article

Montane forests are among the most critical terrestrial biodiversity reservoirs, acting as climate-regulating systems and water towers, and are vital for human well-being across the world. However, they are facing diverse, repeated, and contradictory challenges due to climate variability and change that may undermine their resilience. This study analyzed spatiotemporal trends in temperature and rainfall in the Aberdare Forest ecosystem, Kenya, over the past three decades (1994–2024). ERA5-Land (temperature, 9 km resolution) and CHIRPS-v3 (precipitation, 5 km resolution) provided the gridded climate datasets and were validated against observed station data from the Kenya Meteorological Department (KMD). Trend detection utilized the non-parametric Mann–Kendall (MK) test, while Sen’s slope estimator quantified the magnitude and direction of change. Results show a statistically significant positive trend in temperature, while rainfall demonstrated a statistically non-significant positive trend over the study period. Spatial analysis revealed statistically significant warming across the forest, bamboo, and moorland ecological zones, while rainfall showed positive but non-significant trends, with the magnitude of both temperature and rainfall trends varying across zones. These findings provide robust empirical evidence of significant temporal and spatial warming with no clear corresponding directional change in rainfall, underscoring increasing climate pressures on the Aberdare Forest ecosystem. The results establish one of the first empirically validated long-term climate trend baselines for the Aberdare Forest ecosystem. This benchmark provides critical evidence to inform adaptive governance and responsive management for enhanced resilience of East Africa’s montane forests. Full article

Temperate native grasslands in southeastern Australia have been extensively cleared for agriculture, and the remaining patches are under growing pressure from further land use change, climate variability, and invasive species. Mapping and monitoring their distribution and the cover of native and exotic grasses are critical for their conservation and management. Field-based methods are not always scalable or time-effective, and this study aimed to develop a scalable method to map and monitor the fractional cover-class maps of native C3 and native C4 grass cover as a component of remnant native grasslands on the western outskirts of Melbourne, Victoria, Australia. Field-based reference data for training and validation of random forest machine learning models were collected across multiple sites in 2021. Sentinel-2 optical spectral bands and vegetation indices were used as the primary input data, and Sentinel-1 Synthetic Aperture Radar (SAR)-derived Grey Level Co-occurrence Matrix (GLCM) texture metrics were assessed for their capacity to improve the model. Results show that random forest models trained on Sentinel-2 data without GLCM texture information derived from Sentinel-1 SAR data provided a moderate overall accuracy (C3: 59.1%, C4: 78.1%). Class-specific metrics showed that reliability was highest for better represented lower-cover classes, particularly the 6–25% native C3 class and the 0–5% native C4 class, while higher-cover classes were less reliable because of the limited number of training and validation samples. Grass cover fractions were modelled well for sparse to moderate grass cover, but dense grass cover was not modelled accurately, probably due to limited high-cover samples in the training dataset. Model performance was not improved by the inclusion of Sentinel-1 SAR-derived GLCM texture metrics, indicating that C-band VH-polarised SAR is not sensitive to the fine-scale structural heterogeneity that characterises native grassland ecosystems. Sparse native C3 and C4 grasses could be mapped most reliably in the lower-cover classes as a component of grasslands with optical remote sensing, and the method developed here can now be applied to enable evidence-based management of grasslands, biodiversity conservation and the monitoring of grassland composition in the WGR and elsewhere. Higher-resolution structural datasets and more sophisticated machine learning approaches may be required to accurately predict native C3 and C4 grass cover fractions in denser grasslands. Full article

Accurate crop-type mapping is essential for agricultural monitoring, but pixel-based products often suffer from within-field fragmentation, boundary noise, and limited consistency with field management units. This study developed a parcel-constrained crop classification approach using the HLSS30 product from the Harmonized Landsat and Sentinel 2 framework. HLSS30 data preprocessing and parcel-level feature extraction were conducted in Google Earth Engine, and parcel-level spectral, vegetation index, and phenological features were used to train a Random Forest classifier with feature selection and nested stratified cross-validation. A pixel-level classification experiment was used as the baseline for comparison. The HLSS30 time series captured class specific differences in canopy establishment, peak greenness, and senescence. Compared with the pixel-level baseline, parcel-level classification increased overall accuracy from 76.5% to 89.4%, increased Kappa from 0.690 to 0.859, and achieved a macro F1 score of 0.8677. Parcel constraints also reduced salt and pepper noise, improved field-level spatial coherence, and supported reliability interpretation through posterior entropy and parcel internal variability. These results indicate that HLSS30-derived temporal and phenological features, when summarized within reliable crop parcel boundaries, provide an efficient and interpretable basis for regional multi-crop mapping in fragmented agricultural landscapes. Full article

Understanding the spatial distribution patterns of pelagic species such as yellowfin tuna (Thunnus albacares) is essential for ecosystem-based fisheries management. However, characterizing CPUE–environment relationships remain challenging because these relationships may be nonlinear and spatially heterogeneous across large oceanic regions. To address these challenges, we developed an interpretable spatial modeling framework, geographically neural network weighted regression integrated with GeoShapley analysis (GNNWR-GeoShapley), which combines the nonlinear learning capability of neural networks with spatially explicit characterization and interpretation of model relationships. Using Pacific longline fishery data and multi-source environmental variables from 2004 to 2023, we constructed quarterly models of CPUE–environment relationships and compared the performance of GNNWR with Generalized additive model (GAM), geographically weighted regression (GWR), graph neural network (GNN) models, and Geographical Random Forest (GRF). The results demonstrated that GNNWR showed the best overall performance across seasons, effectively capturing nonlinear relationships and spatial heterogeneity in yellowfin tuna nominal CPUE. GeoShapley analysis further revealed that sea surface and subsurface (150 m) temperature and salinity were among the most important environmental variables associated with nominal CPUE variations. Nonlinear response patterns indicated that SST values above approximately 25 °C and T150 values above approximately 19 °C were associated with positive model contributions, whereas higher salinity values (>35) exhibited negative contributions. Moreover, spatial effects represented by the geographical location variable (GEO) and their interactions with environmental variables revealed pronounced spatial heterogeneity, with the contribution patterns of environmental factors varying across seasons and regions. This study provides an interpretable spatial modeling framework for characterizing complex species–environment relationships and offers new insights into the spatial variability of Pacific yellowfin tuna nominal CPUE for fisheries oceanography and sustainable resource management. Full article

Flood prediction remains a critical challenge in environmental risk management and disaster preparedness. Accurate river-level forecasting is essential for the development of reliable early warning systems and the mitigation of flood-related risks. However, conventional ensemble approaches, such as averaging and majority voting, often exhibit limited adaptability when individual models respond differently to anomalies or incomplete data. To address this limitation, this study proposes a graph-based synchronization framework that integrates XGBoost and Random Forest models using a Graph Convolutional Network (GCN). The proposed framework represents the outputs of the base prediction models as graph nodes and employs graph message passing to learn context-dependent relationships between their predictions. The framework is evaluated using real-world hydrological observations from the Lepenec River Basin in North Macedonia together with meteorological data obtained from the OpenWeatherMap API. Experimental results demonstrate that the proposed GCN-based synchronization framework outperforms both the standalone prediction models and the previously proposed linear synchronization method, achieving an R² value of 0.91 and a Mean Absolute Error (MAE) of 0.21. The obtained results indicate that graph-based synchronization provides an adaptive approach for integrating heterogeneous machine-learning models and has the potential to support future flood early-warning systems and intelligent environmental monitoring applications. Full article

Artificial intelligence is increasingly used in asset management, although evidence that greater model complexity consistently improves investment performance remains limited. This study examines whether machine-learning methods generate incremental value in European equity selection beyond transparent investment rules. The analysis used 562 eligible monthly price series drawn from the March 2026 STOXX Europe 600 constituents and applied retrospectively as a common ex post reference universe. The models were trained on observations from January 2011 to December 2020 and evaluated out of sample using signals formed from January 2021 to March 2026, with corresponding portfolio returns realized from February 2021 to April 2026. Logistic regression, random forest, XGBoost, and an equal-probability ensemble are compared with momentum, low-volatility, and equal-weight strategies under common portfolio-construction rules. Logistic regression recorded the strongest classification results and the highest gross cumulative portfolio return. The transparent momentum strategy recorded the highest observed Sharpe ratio and substantially lower target-weight turnover, while the three-model ensemble underperformed both logistic regression and momentum. Newey–West inference did not establish a statistically significant difference in mean monthly returns between logistic regression and momentum. The results indicate that additional algorithmic complexity did not generate incremental investment value under the restricted technical information set and portfolio framework considered. The findings support evaluating model complexity jointly through predictive quality, gross performance, statistical uncertainty, implementation sensitivity, interpretability, and governance requirements. Full article

Recent studies on users of urban parks call for the need for more insight into the importance of green spaces as catalysts of more inclusive spaces. This paper contributes to the ongoing debate by examining urban parks as age-inclusive infrastructures in a post-socialist context. Three parks in the City of Timișoara, Romania, are selected as a comparative case study. Our analysis combines systematic field observations and GIS-based mapping of park accessibility and facilities with 42 semi-structured interviews with young, mid-aged and older visitors to the three contrasting parks: a renovated historic central park, a peripheral forest-like park and a small neighbourhood park embedded in a communist-era housing estate. The findings suggest that inclusiveness for all generational categories, as well as attachment for neighbourhood parks, are important drivers for urban parks users. Inclusiveness is driven not only by amenities, but also by park-specific attachment, as well as the quality of maintenance and lighting. These factors shape perceived equity and convenience. We advocate for a differentiated management model tailored to each park, balancing conservation-oriented quiet zones with flexible, event-capable areas. This model prioritises lighting, seating ergonomics and safety measures as core components of age-inclusive planning. Our findings support the need for different management of urban green spaces that capitalises on the social and spatial specificities of each park. The comparison further shows that the smallest and least equipped park generates the strongest attachment and the most regular use across all generations: proximity and continuity of maintenance matter more than surface area or scale of investment. Full article

Regional identification and risk mapping of landslide and debris flow hazard chains (LDHCs) are essential for disaster prevention and land-use planning in mountainous regions. This study proposes a machine learning framework for the regional identification and risk mapping of LDHCs in southern Gansu Province, China. A spatial database was established using an inventory of 160 landslide hazard chains (LHCs) and 121 debris-flow hazard chains (DHCs), together with eight environmental conditioning factors. Considering the distinct geomorphological characteristics of the two hazard types, grid-based and watershed-based mapping units were adopted for landslide and debris-flow susceptibility mapping, respectively. Five ensemble learning algorithms were evaluated to identify the optimal susceptibility models. Hazard maps were subsequently generated by integrating susceptibility with earthquake- and extreme rainfall-triggering factors, and regional risk maps were produced by incorporating population and Gross Domestic Product (GDP) exposure data. The Random Forest model achieved the best performance for LHC susceptibility mapping, with an accuracy of 84.9% and an AUC of 0.914, whereas the Extra Trees model performed best for DHC susceptibility mapping, with an accuracy of 92.6% and an AUC of 0.978. The resulting risk maps indicate that the middle and lower reaches of the Bailong River, including Wudu, Zhouqu, Qin’an, and Tianshui, represent the highest-risk areas for LDHCs. The proposed framework provides an effective and transferable approach for regional identification and risk mapping of LDHCs, offering valuable support for disaster prevention, emergency planning, and land-use management in mountainous regions. Full article

Wildland–urban interfaces (WUIs) are critical zones for wildfire prevention in Mediterranean and semi-arid landscapes, particularly where human settlements interact with continuous combustible vegetation. In Algeria, operational tools for delineating WUI configurations and spatially prioritizing preventive actions remain limited. This study develops a GIS-based structural wildfire-risk prioritization framework for the Djebel El Ouahch massif (Constantine, northeastern Algeria) by combining WUIMAP II settlement typologies with an Aggregation Index (AI) describing the horizontal continuity of arboreal and shrub vegetation derived from Sentinel−2 data. Built-up structures were classified as isolated housing, dispersed housing, main urbanized areas, or peripheral zones and combined with three AI classes to delineate relative structural-priority areas. Dispersed and isolated housing accounted for 55% and 37% of mapped housing structures, respectively, while the WUI contained an estimated 18,100 inhabitants distributed across 3620 housing units. The resulting WUI × AI structural-priority layer covered 0.863 km², of which high- and very-high-priority zones jointly represented 0.544 km² (63.04%). These priority areas identify locations where structurally exposed settlement configurations coincide with greater horizontal fuel continuity, providing a spatial basis for targeted fuel management, surveillance planning, emergency-access improvement, and wildfire-prevention actions. By improving the spatial targeting of preventive measures, the framework can contribute to sustainable forest and land management, disaster-risk reduction, ecosystem protection, and the resilience of WUI settlements. Full article

Water-distribution-system failure records often contain nonlinear, heterogeneous patterns that are difficult to summarize with conventional statistics alone. This study evaluates how Random Forest and XGBoost reconstruct normalized water-pipe failure-rate patterns in a field-derived Malatya, Türkiye dataset and uses distributional diagnostics, empirical ranking, and multi-method explainability to identify stable model drivers. The analysis includes 1231 records from nine source-defined clusters and four modelling inputs: cluster identifier, normalized pipe length, normalized pipe diameter, and normalized second-failure age. Because the models were fitted to and evaluated on the same records, the reported performance represents apparent/full-data goodness of fit rather than independent predictive validation. XGBoost provided the closer reconstruction (R² = 0.9189 versus 0.8523 for Random Forest; MAE = 0.0230 versus 0.0321). Across permutation importance, SHAP attribution, and variable-removal sensitivity, pipe length and second-failure age emerged as the most robust model-relevant features, whereas native importance was more method-dependent. The framework demonstrates how ensemble models can support transparent retrospective database screening and pattern interpretation while maintaining a clear boundary between historical fit and forecasting. Temporal, spatial, pipe-grouped, or external validation is required before prospective decision use. Full article

Reliable assessments of the operational conditions of LED lighting systems are important for maintenance planning and asset management. Conventional Wear Index (WI) approaches typically assume a fixed daily number of switching cycles, which may not accurately reflect actual operating conditions. This study proposes a machine learning-based framework for operational Wear Index estimation using measured switching-cycle data collected from an LED lighting installation. Daily energy consumption and operating time were predicted using Multiple Linear Regression (MLR), Random Forest (RF), Long Short-Term Memory (LSTM), and Backpropagation Neural Network (BPNN) models. The models were evaluated using consistent performance metrics, and the BPNN achieved the highest predictive accuracy and was subsequently applied to the Wear Index calculation. Incorporating measured daily switching activity instead of the conventional fixed assumption of nine switching cycles per day resulted in a 17.80% higher cumulative Wear Index. This result demonstrates that the representation of switching activity can materially influence operational wear assessment. The proposed framework provides a practical data-driven approach for estimating operational wear under actual operating conditions and may support maintenance planning in intelligent lighting systems. Since the results are based on a single monitored installation, they should be interpreted as a case study. Full article

Accurate calibration of Manning’s roughness coefficients is essential for reliable river hydraulic modelling, flood prediction, and water resources management, yet conventional calibration methods often struggle with high-dimensional parameter spaces and nonlinear hydraulic interactions. This study proposes and evaluates a hierarchical artificial intelligence framework for the inverse calibration of spatially distributed Manning’s roughness coefficients across three channel zones (left bank, main channel, and right bank), using a 48 km reach of the Tigris River in Baghdad as a case study. A one-dimensional HEC-RAS hydraulic model based on 30 measured cross-sections generated 18,360 simulations by systematically varying Manning’s roughness coefficients (0.02–0.045). Three calibration strategies were evaluated: (i) a simple Gradient Boosting Regression model based on a weighted composite roughness formula, (ii) conventional machine learning models (Random Forest, Gradient Boosting, and Multi-Layer Perceptron), and (iii) a deep learning framework combining a three-layer neural network (64 → 32 → 16 neurons), Differential Evolution optimisation, and cubic spline interpolation. Calibration accuracy increased with model complexity. The deep learning framework achieved the best performance, reducing the root mean square error by 96.6% (from 1.202 to 0.041 m), with R² = 0.992 and negligible bias (−0.004 m). Conventional machine learning models produced spatially variable Manning’s roughness distributions, with the calibrated main-channel roughness (mean n = 0.0512) being 34.0–57.5% higher than the corresponding bank values. The proposed framework provides an effective approach for calibrating spatially distributed roughness coefficients in one-dimensional hydraulic models, with strong potential to improve river hydraulic simulations and support future applications to flood modelling. Full article

Reliable prediction and interpretation of photosynthetic rate are important for precision environmental management in greenhouse tomato production, but model stability is often limited by variable redundancy, measurement-session effects, and environmental heterogeneity. This study developed an integrated gas-exchange-data-based framework for key-factor selection, photosynthetic-rate reconstruction, cross-session validation, environmental correction, and physiology-informed limitation screening. Core predictors were first identified from high-dimensional gas-exchange variables using K-means clustering and random-forest importance analysis. A CatBoost–ExtraTrees–RBF-SVR stacked ensemble was then constructed to reconstruct the leaf photosynthetic rate from selected gas-exchange variables, and its cross-session generalization was evaluated using nested cross-validation and leave-one-file-out (LOFO) extrapolation. Environmental correction was further applied to improve cross-session comparability, and rule-based limitation screening was used to classify potential environmental constraints associated with reduced photosynthesis. The stacked model achieved an RMSE of 0.806 and an $R^2$ of 0.9918 under nested cross-validation, and an RMSE of 0.814 and an $R^2$ of 0.9917 under LOFO extrapolation. After environmental correction, the cross-session variability of the environmentally corrected photosynthetic indicator was reduced by 96.0%, reflecting improved cross-session comparability. Under deployable sensor inputs, performance decreased markedly (A1: RMSE = 4.578, $R^2$ = 0.735; A2: RMSE = 4.610, $R^2$ = 0.731), compared with the gas-exchange baseline (RMSE = 0.833, $R^2$ = 0.991). Thus, routine greenhouse sensor models should be regarded as approximate rather than high-accuracy substitutes for the gas-exchange-based model. Full article