Multivariate Adaptive Regression Splines Research Articles

Prediction of 6-month poststroke spasticity in patients with acute first-ever stroke by machine learning.

Poststroke spasticity (PSS) reduces arm function and leads to low levels of independence. This study suggested applying machine learning (ML) from routinely available data to support the clinical management of PSS. 172 patients with acute first-ever stroke were included in this prospective cohort study. Twenty clinical information and rehabilitation assessments were obtained to train various ML algorithms for predicting 6-month PSS defined by a modified Ashworth scale (MAS) score ≥ 1. Factors significantly relevant were also defined. The study results indicated that multivariate adaptive regression spline (area under the curve (AUC) value: 0.916; 95% confidence interval (CI): 0.906-0.923), adaptive boosting (AUC: 0.962; 95% CI: 0.952-0.973), random forest (RF) (AUC: 0.975; 95% CI: 0.968-0.981), support vector machine (SVM) (AUC: 0.980; 95% CI: 0.970-0.989) outperformed the traditional logistic model (AUC: 0.897; 95% CI: 0.884-0.910) (P < 0.05). Among all of the algorithms, the RF and SVM models outperformed the others (P < 0.05). FMA score, days in hospital, age, stroke location, and paretic side were the most important features. These findings suggest that ML algorithms can help augment clinical decision-making processes for the assessment of PSS occurrence, which may enhance the efficacy of management for patients with PSS in the future.

Coupling FELA and computation intelligent in predicting behaviour of inclination strip anchor in anisotropic and heterogeneous clays

This study introduces new numerical investigations of the uplift capacity of inclined strip anchors buried in anisotropic and heterogeneous clays. The analyses utilize Lower Bound (LB) Finite Element Limit Analysis techniques (FELA) under plane strain conditions, employing the Anisotropic Undrained Shear (AUS) failure criterion. The undrained shear strength of the clay increases proportionally with depth, and anisotropy is introduced by adjusting the ratios between the shear strengths obtained from three different tests. Four input parameters, namely, the embedment ratio (H/B), the inclination angle (α), the ratio of increasing undrained shear strength with depth (m = ρΒ/suTC0), and the factor of anisotropic parameters (re), are considered in analyzing the uplift coefficient (Fc), and a set of design charts is proposed. The effect of dimensionless parameters on the development of failure mechanisms is also investigated. Moreover, a novel method for assessing the ultimate uplift capacity of an inclined strip anchor in natural clay is given by integrating FELA with Multivariate Adaptive Regression Splines (MARS) and Extreme Gradient Boosting (XGBoost). Based on the optimal machine-learning models, this study also calculates the contribution of each input to the value of Fc via sensitivity analysis. The engineering computations for strip anchors embedded in anisotropic and non-homogenous clays can be improved with the results of these investigations and the proposed machine-learning models.

Aquatic vegetation mapping with UAS-cameras considering phenotypes

Aquatic vegetation species at the genus level in an oxbow lake were identified in Hungary based on a multispectral Uncrewed Aerial System (UAS) survey within an elongated oxbow lake area of the Tisza River under continental climate. Seven and 13 classes were discriminated using three different classification methods (Support Vector Machine [SVM], Random Forest [RF], and Multivariate Adaptive Regression Splines [MARS]) using different input data in ten combinations: original spectral bands, spectral indices, Digital Surface Model (DSM), and Haralick texture indices. We achieved a high (97.1%) overall accuracies (OAs) by applying the SVM classifier, but the RF performed only <1% worse, as it was represented in the first places of the classification rank before the MARS. The highest classification accuracies (>84% OA) were obtained using the most important variables derived by the Recursive Feature Elimination (RFE) method. The best classification required DSM as an input variable. The poorest classification performance belonged to the model that used only texture indices or spectral indices. On the class level, Stratiotes aloides exhibit the lowest degree of separability compared to the other classes. Accordingly, we recommend using supplementary input data for the classifications besides the original spectral bands, for example, DSM, spectral, and texture indices, as these variables significantly improve the classification accuracies in the proper combinations of the input variables.

A framework for estimating the matric suction in unsaturated soils using multiple artificial intelligence techniques

AbstractImplementation of the state‐of‐the‐art understanding of the mechanics of unsaturated soils into geotechnical engineering practice is partly limited due to the lack of quick, reliable, and economical techniques for matric suction measurement. Matric suction is one of the key stress state variables that significantly influences the hydro‐mechanical behavior of unsaturated soils. In this paper, to address this objective, two artificial intelligence (AI) models were developed for estimating matric suction in unsaturated soils based on the particle swarm optimization support vector regression (PSO‐SVR) and multivariate adaptive regression spline (MARS) algorithms. The results suggest that both these models can reasonably estimate matric suction. Compared to the MARS model, the PSO‐SVR model can achieve higher accuracy. Nonetheless, the MARS model facilitates the sensitivity analysis and the selection of essential inputs. A novel integrated framework is proposed and validated, leveraging the strengths, and alleviating the limitations of the PSO‐SVR and MARS algorithms for reliable and rapid estimation of matric suction in the range of 0–1500 kPa for low plastic soils (0 &lt; Ip ≤ 7). Six inputs are required to use this model successfully; some can be measured using conventional laboratory tests, and others can be calculated from mass‐volume relationships.

-A machine learning model to predict surgical site infection after surgery of lower extremity fractures.

This study aimed to develop machine learning algorithms for identifying predictive factors associated with the risk of postoperative surgical site infection in patients with lower extremity fractures. A machine learning analysis was conducted on a dataset comprising 1,579 patients who underwent surgical fixation for lower extremity fractures to create a predictive model for risk stratification of postoperative surgical site infection. We evaluated different clinical and demographic variables to train four machine learning models (neural networks, boosted generalised linear model, naïve bayes, and penalised discriminant analysis). Performance was measured by the area under the curve score, Youdon's index and Brier score. A multivariate adaptive regression splines (MARS) was used to optimise predictor selection. The final model consisted of five predictors. (1) Operating room time, (2) ankle region, (3) open injury, (4) body mass index, and (5) age. The best-performing machine learning algorithm demonstrated a promising predictive performance, with an area under the ROC curve, Youdon's index, and Brier score of 77.8%, 62.5%, and 5.1%-5.6%, respectively. The proposed predictive model not only assists surgeons in determining high-risk factors for surgical site infections but also empowers patients to closely monitor these factors and take proactive measures to prevent complications. Furthermore, by considering the identified predictors, this model can serve as a reference for implementing preventive measures and reducing postoperative complications, ultimately enhancing patient outcomes. However, further investigations involving larger datasets and external validations are required to confirm the reliability and applicability of our model.

Risk factors and prediction model for new-onset hypertensive disorders of pregnancy: a retrospective cohort study

Risk factors and prediction model for new-onset hypertensive disorders of pregnancy: a retrospective cohort study

Machine learning algorithms for the prediction of drought conditions in the Wami River sub-catchment, Tanzania

Study regionThis study refers to the Wami river sub-catchments in Eastern Tanzania. Study FocusThe five-machine learning (ML) algorithms, including long short-term memory (LSTM), multivariate adaptive regression spline (MARS), support vector machine (SVM), extreme learning machine (ELM), and M5 Tree, were used to predict the most widely used drought index, the standard precipitation index (SPI), at six and nine months. Algorithms were established using monthly rainfall data for the period from 1990 to 2022 at five meteorological stations distributed across the Wami River sub-catchment: Barega, Dakawa, Dodoma, Kongwa, and Mandera stations. New hydrological insights for the region. The predicted results of all five ML algorithms were evaluated using several statistical metrics, including Pearson’s correlation coefficient (R), mean absolute error (MAE), root mean square error (RMSE), and Nash Sutcliffe efficiency (NSE). The prediction results revealed that LSTM perform better in predicting drought conditions using SPI6 (6-month SPI) and SPI9 (9-month SPI) with the highest NSE of 0.99 in all five stations, and R of 0.99 in four stations except at Kongwa station, where R range from 0.75 to 0.99. These prediction results will aid decision-makers and planners to develop a drought monitoring and drought early warning system in order to strengthen the governance and resilience to the catchment and people on the impacts of water scarcity and climate change.

Development of multiple explicit data-driven models for accurate prediction of CO2 minimum miscibility pressure

This study presents utilization of multiple data-driven models for predicting CO2 minimum miscibility pressure (MMP). The aim is to address the issue of existing models lacking explicit presentation. With a database of 155 data points, five models were developed using artificial neural network (ANN), multigene genetic programming (MGGP), support vector regression (SVR), multivariate adaptive regression splines (MARS), and multiple linear regression (MLR). Comparative analysis was conducted using statistical metrics (R2, MSE, MAE, RMSE), and sensitivity analysis was performed on input variables. The results showed that ANN and SVR had comparable predictive performance (ANN: R2 = 0.982, MSE = 0.00676, MAE = 0.9765, RMSE = 0.082), SVR (R2 = 0.935, MSE = 0.0041, MAE = 0.72, RMSE = 0.064) followed by MARS, MLR, and MGGP. Sensitivity analysis revealed that reservoir temperature was the most influential parameter across all models, except for the MLR algorithm where injected CO2 amount was crucial. These models can be used for a wide range of CO2 MMP ranging from 940 psi to 5830 psi, thus rendering them useful for any reservoir globally. These models offer improved accuracy and computational efficiency compared to existing ones, potentially reducing costs associated with laboratory experiments and providing rapid and precise CO2 MMP predictions.

Trends and amount changes of temperature and precipitation under future projections in high–low groups and intra-period for the Eastern Black Sea, the Wettest Basin in Türkiye

AbstractThis study investigates the possible effects of climate change on temperature and precipitation variables in the Eastern Black Sea Basin, Türkiye’s wettest and flood-prone region. The outputs of three GCMs under historical, RCP4.5, and RCP8.5 scenarios were downscaled to regional scale using the multivariate adaptive regression splines method. The future monthly temperature and precipitation for 12 stations in the basin were projected for three periods: the 2030s (2021–2050), 2060s (2051–2080), and 2090s (2081–2100). In addition to relative changes, high and low groups and intra-period trends were analyzed for the first time using innovative methods. For the pessimistic scenario, an increase of 3.5 °C in the interior and 3.0 °C in the coastal areas of the basin is projected. For the optimistic scenario, these values are expected to be 2.5 and 2.0 °C, respectively. A decrease in precipitation is projected for the interior region, and a significant increase is expected for the eastern and coastal areas of the basin, especially in spring. This result indicates that floods will occur frequently coastal areas of the basin in the coming periods. Also, although the monotonic trends of temperatures during periods are higher than precipitation in interior regions, these regions may have more uncertainty as their trends are in different directions of low and high groups of different scenarios and GCMs and contribute to all trends, especially precipitation.

Strength assessment of structural masonry walls: analysis based on machine learning approaches

ABSTRACT In conventional masonry buildings, masonry walls are key structural load-bearing elements. Likewise, masonry infill walls strengthen framed constructions against lateral stress. The material characteristics of brick units and mortar determine the compressive strength of structural masonry walls. In this study, advanced machine learning (ML) techniques were utilized to estimate the compressive strength of structural masonry walls based on the material properties of brick units and mortar. The Young’s modulus of brick units (Eu), compressive strength of brick units (Fcu), Young’s modulus of mortar (Em), and compressive strength of mortar (Fcm) were used as input parameters to model the compressive strength (Fc) of the structural masonry wall. Gradient Tree Boosting (GTB), Elman Neural Network (ENN), and Multivariate Adaptive Regression Splines (MARS) were developed using four diverse input and output (I/O) combinations to explore the effect of each input parameter on output estimation. The data used for modeling were obtained from prior studies published in the literature. The model’s performance was evaluated based on different statistical (error and efficiency) indices. For the third and fourth (I/O) combinations, the MARS model significantly outperformed other models. However, for the first I/O combination, the GTB model performed well. The study also revealed that the compressive strength of a structural masonry wall is more likely to depend on the strength and quality of the brick units than on the strength of the mortar.

Potential distribution of Biscogniauxia mediterranea and Obolarina persica causal agents of oak charcoal disease in Iran’s Zagros forests

In Iran, native oak species are under threat from episodes of Charcoal Disease, a decline syndrome driven by abiotic stressors (e.g. drought, elevated temperature) and biotic components, Biscogniauxia mediterranea (De Not.) Kuntze and Obolarina persica (M. Mirabolfathy). The outbreak is still ongoing and the country’s largest ever recorded. Still, the factors driving its’ epidemiology in time and space are poorly known and such knowledge is urgently needed to develop strategies to counteract the adverse effects. In this study, we developed a generic framework based on experimental, machine-learning algorithms and spatial analyses for landscape-level prediction of oak charcoal disease outbreaks. Extensive field surveys were conducted during 2013–2015 in eight provinces (more than 50 unique counties) in the Zagros ecoregion. Pathogenic fungi were isolated and characterized through morphological and molecular approaches, and their pathogenicity was assessed under controlled water stress regimes in the greenhouse. Further, we evaluated a set of 29 bioclimatic, environmental, and host layers in modeling for disease incidence data using four well-known machine learning algorithms including the Generalized Linear Model, Gradient Boosting Model, Random Forest model (RF), and Multivariate Adaptive Regression Splines implemented in MaxEnt software. Model validation statistics [Area Under the Curve (AUC), True Skill Statistics (TSS)], and Kappa index were used to evaluate the accuracy of each model. Models with a TSS above 0.65 were used to prepare an ensemble model. The results showed that among the different climate variables, precipitation and temperature (Bio18, Bio7, Bio8, and bio9) in the case of O. persica and similarly, gsl (growing season length TREELIM, highlighting the warming climate and the endophytic/pathogenic nature of the fungus) and precipitation in case of B. mediterranea are the most important influencing variables in disease modeling, while near-surface wind speed (sfcwind) is the least important variant. The RF algorithm generates the most robust predictions (ROC of 0.95; TSS of 0.77 and 0.79 for MP and OP, respectively). Theoretical analysis shows that the ensemble model (ROC of 0.95 and 0.96; TSS = 0.79 and 0.81 for MP and OP, respectively), can efficiently be used in the prediction of the charcoal disease spatiotemporal distribution. The oak mortality varied ranging from 2 to 14%. Wood-boring beetles association with diseased trees was determined at 20%. Results showed that water deficiency is a crucial component of the oak decline phenomenon in Iran. The Northern Zagros forests (Ilam, Lorestan, and Kermanshah provinces) along with the southern Zagros forests (Fars and Kohgilouyeh va-Boyer Ahmad provinces) among others are the most endangered areas of potential future pandemics of charcoal disease. Our findings will significantly improve our understanding of the current situation of the disease to pave the way against pathogenic agents in Iran.

BMB-LIME: LIME with modeling local nonlinearity and uncertainty in explainability

The majority of eXplainable Artificial Intelligence (XAI) methods assume the local linearity of the decision boundary, leading to significant errors when dealing with non-linear local decision boundaries. Moreover, explanation methods relying on perturbation samples yield only a constant value for feature contributions. This study aims to overcome these limitations by introducing a novel explanation method that accurately captures non-linear local classification boundaries or regression splines while also quantifying the uncertainty associated with each feature contribution. Expanding upon the state-of-the-art XAI method, Local Interpretable Model-agnostic Explanations (LIME), we propose a nonlinear local explainer called BMB-LIME (Bootstrap aggregating Multivariate adaptive regression splines Bayesian LIME). BMB-LIME is constructed using weighted multivariate adaptive regression splines (MARS) with bootstrap aggregating and models the uncertainty of feature contributions within a Bayesian framework. Through a series of experiments, we demonstrate the superior performance of the BMB-LIME local explainer over baseline methods in terms of local fidelity and stability. Its adjusted r-squared is statistically significant by at least 80.5 percent, and it consistently demonstrates over 89.1 percent consistency in feature contribution and Jaccard similarity across diverse real-world and simulated datasets. The proposed method not only enhances fidelity and stability but also improves the measurement of uncertainty in explanations, thereby significantly contributing to the trustworthiness and transparency of XAI.

Ensembling machine learning models to identify forest fire-susceptible zones in Northeast India

Forest fires pose significant challenges by disrupting ecological balance, impacting socio-economic harmony, and raising global concerns. North-East India (NEI) experiences high incidences of forest fires, making it crucial to implement suitable management measures considering the driving forces influencing fire likelihood. This study aims to identify forest fire susceptibility zones in NEI by using five machine-learning modeling approaches, Boosted Regression Tree (BRT), Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Tree (CART), and Multivariate Adaptive Regression Splines (MARS), and an ensemble method. Forest fire data from the SNPP – VIIRS sensor (2018–2019) were rectified for spatial autocorrelation. Thirty-two responsive predictor variables related to topographic, climatic, biophysical, and anthropogenic factors were used as model inputs and multicollinearity analysis was performed to eliminate highly correlated predictors. Results indicate that the southern and southeastern regions of NEI, characterized by ample solar radiation, enhanced vegetation index, high human population density, and jhum cultivation, contribute significantly to higher susceptibility to forest fires. The Random Forest model performs best among the models employed, achieving an AUC value of 0.87. The ensemble susceptibility map, binarized based on AUC weighting, covers 29.54% of the total geographic area and 44.42% of the forested area of NEI. The vulnerability levels vary among states, with Mizoram showing the highest susceptibility at 89.27% and Sikkim exhibiting the lowest vulnerability at only 0.49% of their respective geographic areas. This map provides valuable insights for implementing effective forest fire management plans in the region. Moreover, the methodology utilized in this study, which incorporates satellite imagery, GIS techniques, and improved modeling techniques, can be replicated in any geographical region worldwide to facilitate effective forest fire management at a regional to large scale.

Comparison of Predictive Performance of Data Mining Algorithms in Predicting Tomato Yield with the A Case Study in Igdir

Among the vegetable species in the world, the plant with the most cultivation area is tomato. Increasing tomato yield is important in terms of contributing more to the world economy, producer’s income and human health. With the advancement in software technologies, the importance of data mining algorithms is increasing due to the fact that these algorithms can produce more sophisticated solutions for regression and classification problems. Determining the factors affecting tomato yield and comparing different data mining algorithms on prediction of tomato yield are the purpose of this study. For this purpose, survey study was conducted with the 105 farmers, selected by Simple Random Sampling Method in Igdir province in 2016. Different data mining algorithms including Classification and Regression Tree, Exhaustive CHAID, Chi-Square Automatic Interaction Detector, Artificial Neural Network Algorithm, Multivariate Adaptive Regression Splines and General Linear Model were developed and compared their predictive performance. MARS decision tree has been able to build a model with greatest predictive accuracy, and the others are respectively ANN, GLM, CART, CHAID and Exhaustive CHAID. In the MARS model, number of irrigation , amount of chemical fertilizer , age of farmer , number of seedlings , education level , soil analysis status , sowing region were found statistically significant (P˂0.05). Preferring the MARS model could give an opportunity to detect factors affecting tomato yield and their interactions with higher accuracy. Moreover, results can be easily interpreted and the rules are understandable.

Optimal Implant Sizing Using Machine Learning Is Associated With Increased Range of Motion After Cervical Disk Arthroplasty.

Cervical disk arthroplasty (CDA) offers the advantage of motion preservation in the treatment of focal cervical pathology. At present, implant sizing is performed using subjective tactile feedback and imaging of trial cages. This study aims to construct interpretable machine learning (IML) models to accurately predict postoperative range of motion (ROM) and identify the optimal implant sizes that maximize ROM in patients undergoing CDA. Adult patients who underwent CDA for single-level disease from 2012 to 2020 were identified. Patient demographics, comorbidities, and outcomes were collected, including symptoms, examination findings, subsidence, and reoperation. Affected disk height, healthy rostral disk height, and implant height were collected at sequential time points. Linear regression and IML models, including bagged regression tree, bagged multivariate adaptive regression spline, and k-nearest neighbors, were used to predict ROM change. Model performance was assessed by calculating the root mean square error (RMSE) between predicted and actual changes in ROM in the validation cohort. Variable importance was assessed using RMSE loss. Area under the curve analyses were performed to identify the ideal implant size cutoffs in predicting improved ROM. Forty-seven patients were included. The average RMSE between predicted and actual ROM was 7.6° (range: 5.8-10.1) in the k-nearest neighbors model, 7.8° (range: 6.5-10.0) in the bagged regression tree model, 7.8° (range: 6.2-10.0) in the bagged multivariate adaptive regression spline model, and 15.8° (range: 14.3-17.5°) in a linear regression model. In the highest-performing IML model, graft size was the most important predictor with RMSE loss of 6.2, followed by age (RMSE loss = 5.9) and preoperative caudal disk height (RMSE loss = 5.8). Implant size at 110% of the normal adjacent disk height was the optimal cutoff associated with improved ROM. IML models can reliably predict change in ROM after CDA within an average of 7.6 degrees of error. Implants sized comparably with the healthy adjacent disk may maximize ROM.

Early detection of nasopharyngeal carcinoma through machine-learning-driven prediction model in a population-based healthcare record database.

Early diagnosis and treatment of nasopharyngeal carcinoma (NPC) are vital for a better prognosis. Still, because of obscure anatomical sites and insidious symptoms, nearly 80% of patients with NPC are diagnosed at a late stage. This study aimed to validate a machine learning (ML) model utilizing symptom-related diagnoses and procedures in medical records to predict nasopharyngeal carcinoma (NPC) occurrence and reduce the prediagnostic period. Data from a population-based health insurance database (2001-2008) were analyzed, comparing adults with and without newly diagnosed NPC. Medical records from 90 to 360 days before diagnosis were examined. Five ML algorithms (Light Gradient Boosting Machine [LGB], eXtreme Gradient Boosting [XGB], Multivariate Adaptive Regression Splines [MARS], Random Forest [RF], and Logistics Regression [LG]) were evaluated for optimal early NPC detection. We further use a real-world data of 1 million individuals randomly selected for testing the final model. Model performance was assessed using AUROC. Shapley values identified significant contributing variables. LGB showed maximum predictive power using 14 features and 90 days before diagnosis. The LGB models achieved AUROC, specificity, and sensitivity were 0.83, 0.81, and 0.64 for the test dataset, respectively. The LGB-driven NPC predictive tool effectively differentiated patients into high-risk and low-risk groups (hazard ratio: 5.85; 95% CI: 4.75-7.21). The model-layering effect is valid. ML approaches using electronic medical records accurately predicted NPC occurrence. The risk prediction model serves as a low-cost digital screening tool, offering rapid medical decision support to shorten prediagnostic periods. Timely referral is crucial for high-risk patients identified by the model.

Modeling of kappa factor using multivariate adaptive regression splines: application to the western Türkiye ground motion dataset

AbstractThe recent seismic activity on Türkiye’s west coast, especially in the Aegean Sea region, shows that this region requires further attention. The region has significant seismic hazards because of its location in an active tectonic regime of North–South extension with multiple basin structures on soft soil deposits. Recently, despite being 70 km from the earthquake source, the Samos event (with a moment magnitude of 7.0 on October 30, 2020) caused significant localized damage and collapse in the Izmir city center due to a combination of basin effects and structural susceptibility. Despite this activity, research on site characterization and site response modeling, such as local velocity models and kappa estimates, remains sparse in this region. Kappa values display regional characteristics, necessitating the use of local kappa estimations from previous earthquake data in region–specific applications. Kappa estimates are multivariate and incorporate several characteristics such as magnitude and distance. In this study, we assess and predict the trend in mean kappa values using three–component strong–ground motion data from accelerometer sites with known VS30 values throughout western Türkiye. Multiple linear regression (MLR) and multivariate adaptive regression splines (MARS) were used to build the prediction models. The effects of epicentral distance Repi, magnitude Mw, and site class (VS30) were investigated, and the contributions of each parameter were examined using a large dataset containing recent seismic activity. The models were evaluated using well–known statistical accuracy criteria for kappa assessment. In all performance measures, the MARS model outperforms the MLR model across the selected sites.

Fair Multivariate Adaptive Regression Splines for Ensuring Equity and Transparency

Predictive analytics has been widely used in various domains, including education, to inform decision-making and improve outcomes. However, many predictive models are proprietary and inaccessible for evaluation or modification by researchers and practitioners, limiting their accountability and ethical design. Moreover, predictive models are often opaque and incomprehensible to the officials who use them, reducing their trust and utility. Furthermore, predictive models may introduce or exacerbate bias and inequity, as they have done in many sectors of society. Therefore, there is a need for transparent, interpretable, and fair predictive models that can be easily adopted and adapted by different stakeholders. In this paper, we propose a fair predictive model based on multivariate adaptive regression splines (MARS) that incorporates fairness measures in the learning process. MARS is a non-parametric regression model that performs feature selection, handles non-linear relationships, generates interpretable decision rules, and derives optimal splitting criteria on the variables. Specifically, we integrate fairness into the knot optimization algorithm and provide theoretical and empirical evidence of how it results in a fair knot placement. We apply our fairMARS model to real-world data and demonstrate its effectiveness in terms of accuracy and equity. Our paper contributes to the advancement of responsible and ethical predictive analytics for social good.

Modeling liquid rate through wellhead chokes using machine learning techniques

Precise measurement and prediction of the fluid flow rates in production wells are crucial for anticipating the production volume and hydrocarbon recovery and creating a steady and controllable flow regime in such wells. This study suggests two approaches to predict the flow rate through wellhead chokes. The first is a data-driven approach using different methods, namely: Adaptive boosting support vector regression (Adaboost-SVR), multivariate adaptive regression spline (MARS), radial basis function (RBF), and multilayer perceptron (MLP) with three algorithms: Levenberg–Marquardt (LM), bayesian-regularization (BR), and scaled conjugate gradient (SCG). The second is a developed correlation that depends on wellhead pressure (Pwh), gas-to-liquid ratio (GLR), and choke size (Dc). A dataset of 565 data points is available for model development. The performance of the two suggested approaches is compared with earlier correlations. Results revealed that the proposed models outperform the existing ones, with the Adaboost-SVR model showing the best performance with an average absolute percent relative error (AAPRE) of 5.15% and a correlation coefficient of 0.9784. Additionally, the results indicated that the developed correlation resulted in better predictions compared to the earlier ones. Furthermore, a sensitivity analysis of the input variable was also investigated in this study and revealed that the choke size variable had the most significant effect, while the Pwh and GLR showed a slight effect on the liquid rate. Eventually, the leverage approach showed that only 2.1% of the data points were in the suspicious range.

Evaluation of Downward Surface Longwave Flux Estimates Using Meteosat Cloud Observations

AbstractThe downward surface longwave flux (DSLF) plays a relevant role in the Earth’s surface radiative budget, which is crucial to monitor, understand and model the impact of changes at local and global scales on surface temperature and surface conditions. This study focuses on the evaluation and intercomparison of four DSLF products: (a) a recently developed all‐weather DSLF product based on the multivariate adaptive regression splines (MARS) algorithm driven by satellite cloud information from the Meteosat Second Generation (MSG) and ERA5 reanalysis screen variables; (b) the Satellite Application Facility on Land Surface Analysis (LSA SAF); (c) CERES Synoptic top‐of‐atmosphere and surface fluxes and clouds (CERES‐SYN1deg) and (d) ERA5 reanalysis. The study covers the period 2005–2021 and the MSG region focusing on monthly means. The evaluation performed against 48 ground stations from the Baseline Surface Radiation Network (BSRN) and FLUXNET2015 networks showed that the MARS product outperforms the remaining products, particularly the LSA SAF, while ERA5 and CERES show similar performance metrics. The fours products are intercompared in terms of their mean spatial variability and temporal mean annual cycles and inter‐annual variability in four selected regions, showing a high level of agreement, particularly between MARS, ERA5 and CERES. Our results highlight the clear added value of MARS with respect to LSA SAF, while providing higher spatial resolution (0.05°), constrained by satellite cloud information, when compared with ERA5 (0.25°) or CERES (1°).