Spatial modeling of susceptibility to subsidence using machine learning techniques

Analyzing angle crashes at unsignalized intersections using machine learning techniques

This chapter provides a description of how to use PROC ADAPTIVEREG for generating multivariate adaptive regression splines (MARS) models for univariate continuous and dichotomous outcomes as well as how to evaluate and compare MARS models with likelihood cross-validation (LCV) scores. Comparison of MARS models to adaptive fractional polynomial models on the basis of LCV scores is also covered as well as how to adaptively transform MARS models. Example code is provided for generating models for predicting the univariate continuous outcome death rate per 100,000 in terms of available predictors as also addressed in Chaps 2, 3, 16, and 17 as well as models for predicting the univariate dichotomous outcome a high mercury level in fish over 1.0 ppm versus a lower level in terms of available predictors as also addressed in Chaps 8, 9, 16 and 17.

This study compares the predictive accuracy of eight state‐of‐the‐art modelling techniques for 12 landforms types in a cold environment. The methods used are Random Forest (RF), Artificial Neural Networks (ANN), Generalized Boosting Methods (GBM), Generalized Linear Models (GLM), Generalized Additive Models (GAM), Multivariate Adaptive Regression Splines (MARS), Classification Tree Analysis (CTA) and Mixture Discriminant Analysis (MDA). The spatial distributions of 12 periglacial landforms types were recorded in sub‐Arctic landscape of northern Finland in 2032 grid squares at a resolution of 25 ha. First, three topographic variables were implemented into the eight modelling techniques (simple model), and then six other variables were added (three soil and three vegetation variables; complex model) to reflect the environmental conditions of each grid square. The predictive accuracy was measured by two methods: the area under the curve (AUC) of a receiver operating characteristic (ROC) plot, and the Kappa index (κ), based on spatially independent model evaluation data. The mean AUC values of the simple models varied between 0·709 and 0·796, whereas the AUC values of the complex model ranged from 0·725 to 0·825. For both simple and complex models GAM, GLM, ANN and GBM provided the highest predictive performances based on both AUC and κ values. The results encourage further applications of the novel modelling methods in geomorphology. Copyright © 2008 John Wiley & Sons, Ltd.

Surface longwave net radiation (LWNR) is a vital component in the surface radiation budget. Major progress has been made in the estimations of clear-sky LWNR. However, the estimation of cloudy-sky LWNR remains a significant challenge. In this paper, a linear model (LM) and a multivariate adaptive regression spline (MARS) model were developed to estimate the cloudy-sky LWNR from a satellite-derived surface shortwave net radiation product. Spatially and temporally matched satellite data and ground-measured LWNR, which was collected at 24 sites from four networks, were used to build and validate the linear and MARS models. The effects of land cover, climate type, and surface elevation on the estimate of LWNR were also analyzed. The MARS model, incorporating the normalized difference vegetation index (NDVI) and surface elevation (H) as the inputs, had the best performance. The determination coefficient, BIAS, and root mean square error (RMSE) were 0.51, 0.01 W/m2, and 26.10 W/m2, respectively. The developed model, when combined with freely distributed Global LAnd Surface Satellite (GLASS) products, showed promise for producing surface LWNR and all-sky surface net radiation.

This study predicts the strength of rotary brace damper by analyzing a new set of probabilistic models using the usual method of multiple linear regressions (MLR) and advanced machine-learning methods of multivariate adaptive regression splines (MARS), Rotary brace damper can be easily assembled with high energy-dissipation capability. To investigate the behavior of this damper in structures, a steel frame is modeled with this device subjected to monotonic and cyclic loading. Several response parameters are considered, and the performance of damper in reducing each response is evaluated. MLR and MARS methods were used to predict the strength of this damper. Displacement was determined to be the most effective parameter of damper strength, whereas the thickness did not exhibit any effect. Adding thickness parameter as inputs to MARS and MLR models did not increase the accuracies of the models in predicting the strength of this damper. The MARS model with a root mean square error (RMSE) of 0.127 and mean absolute error (MAE) of 0.090 performed better than the MLR model with an RMSE of 0.221 and MAE of 0.181.

Driven pile is widely used as an effective and convenient structural component to transfer superstructure loads to deep stiffer soils. Nevertheless, during the design process of piles, due to the intrinsic complexity as well as various design variables, the internal stress state related to pile drivability remains unclear, which makes the analysis imprecise. Thus, the development of an accurate predictive model becomes emergent. This paper presents a practical approach to assess pile drivability in relation to the prediction of Maximum compressive stresses and Blow per foot using a series of machine learning algorithms. A database of more than 4000 piles is employed to construct random forest regression (RFR) and multivariate adaptive regression splines (MARS) models. The 10-fold cross-validation method and Lasso regularisation are adapted to obtain the model of superior generalisation ability and better persuasive results. Lastly, the results of RFR and MARS models were compared and evaluated in accordance with the goodness of fit, running time and interpretability. The results show that the RFR model performs better than the MARS in terms of fitting and operational efficiency, but is short of interpretability.

Application of least square support vector machine and multivariate adaptive regression spline models in long term prediction of river water pollution

Background and Aims: Visceral adipose tissue (VAT) is a key cardiometabolic risk factor. This study evaluates the association between VAT and adiposity indices and identifies reliable predictors of increased VAT. Methods: This cross-sectional study utilized data from 4696 participants in the National Health and Nutrition Examination Survey 2011-2018. VAT was measured via dual-energy X-ray absorptiometry. Adiposity indices included body mass index (BMI), waist circumference (WC), lipid accumulation product, visceral adiposity index, body shape index, body roundness index, and metabolic score for visceral fat (METS-VF). Correlation analysis, receiver operating characteristic curve analysis, and multivariate adaptive regression splines (MARS) modeling evaluated the performance of indices and identified key predictors of VAT. Results: All adiposity indices were significantly correlated with VAT (P < 0.001). Among them, METS-VF demonstrated the highest predictive performance for increased VAT (>130 cm2) followed by WC. Optimal cutoff values for METS-VF were 7.1 [areas under the curve (AUC): 0.887, 95% confidence interval (CI): 0.873-0.899] in men and 7.5 (AUC: 0.904, 95% CI: 0.891-0.916) in women. For WC, the cutoff values were 99.5 cm (AUC: 0.866, 95% CI: 0.851-0.879) in men and 96 cm (AUC: 0.883, 95% CI: 0.869-0.896) in women. MARS modeling identified race, age, WC, BMI, glucose, high-density lipoprotein cholesterol, and triglycerides as significant predictors of VAT, achieving an R2 of 75.2%. Conclusion: METS-VF demonstrated the highest predictive value among the indices evaluated for predicting increased VAT. It may serve as a valuable tool in assessing visceral obesity and associated cardiometabolic risks.

The precise forecasting of daily solar radiation (DSR) is receiving prominent attention among thriving solar energy studies. In this study, three standalone models, including gene expression programing (GEP), multivariate adaptive regression splines (MARS), and self-adaptive MARS (SaMARS), were evaluated to forecast DSR. A SaMARS model was classified as MARS model when using the crow search algorithm (CSA). In addition, to overcome the limitations of the standalone models, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) was employed to enhance the accuracy of DSR forecasting. Therefore, three hybrid models including CEEMDAN-GEP, CEEMDAN-MARS, and CEEMDAN-SaMARS were proposed to forecast DSR in Busan and Incheon stations in South Korea. The performance of proposed models were evaluated and affirmed that the accuracy of the CEEMDAN-SaMARS model (NSE = 0.878–0.883) outperformed CEEMDAN-MARS (NSE = 0.819–0.818), CEEMDAN-GEP (NSE = 0.873–0.789), SaMARS (NSE = 0.846–0.769), MARS (NSE = 0.819–0.758), and GEP (NSE = 0.814–0.755) models at both stations. Therefore, it can be concluded that the optimized CEEMDAN-SaMARS model significantly enhanced the accuracy of DSR forecasting compared to that of standalone models.

Groundwater salinization poses a critical threat to sustainable development in arid and semi-arid rurbanizing regions, exemplified by Kerman Province, Iran. This region experiences groundwater ecosystem degradation as a result of the rapid conversion of rural agricultural land to urban areas under chronic drought conditions. This study aims to enhance Groundwater Pollution Risk (GwPR) mapping by integrating the DRASTIC index with machine learning (ML) models, including Random Forest (RF), Boosted Regression Trees (BRT), Generalized Linear Model (GLM), Support Vector Machine (SVM), and Multivariate Adaptive Regression Splines (MARS), alongside hydrogeochemical investigations, to promote sustainable water management in Kerman Province. The RF model achieved the highest accuracy with an Area Under the Curve (AUC) of 0.995 in predicting GwPR, outperforming BRT (0.988), SVM (0.977), MARS (0.951), and GLM (0.887). The RF-based map identified new high-vulnerability zones in the northeast and northwest and showed an expanded moderate vulnerability zone, covering 48.46% of the study area. Analysis revealed exceedances of WHO standards for total hardness (TH), sodium, sulfates, chlorides, and electrical conductivity (EC) in these high-vulnerability areas, indicating contamination from mineralized aquifers and unsustainable agricultural practices. The findings underscore the RF model’s effectiveness in groundwater prediction and highlight the need for stricter monitoring and management, including regulating groundwater extraction and improving water use efficiency in riverine aquifers.

Weed infestations have been identified as a major cause of yield reductions in canola (Brassica napus L.), a vital oil crop that has gained significant prominence in Iran, especially within Fars Province. Weed management using machine learning algorithms has become a crucial approach within the framework of precision agriculture for enhancing the efficacy and efficiency of weed control strategies. The evolution of habitat suitability models for weeds represents a significant advancement in agricultural technology, offering the capability to predict weed occurrence and proliferation accurately and reliably. This study focuses on the issue of dominant weed infestation in canola cultivation, particularly emphasizing the prevalence and impact of wild oat (Avena fatua L.) as the dominant weed species in canola farming in 2023. We collected data on 12 environmental variables related to topography, climate, and soil properties to develop habitat suitability models. Three machine learning techniques, including random forest (RF), support vector machine (SVM), and boosted regression tree (BRT), were estimated based on the receiver operating characteristic (ROC) and area under the curve (AUC) to model the distribution of A. fatua. Model performance was quantified using the ROC curve and AUC metrics to identify the best predictive algorithm. The findings indicated that RF, BRT, and SVM models exhibited accuracies of 99%, 97%, and 96% for the habitat suitability of A. fatua, respectively. The Boruta feature selection method identified the slope variable as significantly influential in A. fatua habitat suitability modeling, followed by plan curvature, clay, temperature, and silt. This study serves as a case study that highlights the utility of machine learning for habitat suitability predictions when information on multiple environmental variables is available. This approach supports effective weed management strategies, potentially enhancing canola productivity and mitigating the ecological impacts associated with weed infestation.

Multivariate adaptive regression splines (MARS) in chromatographic quantitative structure–retention relationship studies

Application of Multivariate Adaptive Regression Splines (MARS) approach in prediction of compressive strength of eco-friendly concrete

Spatial mapping Zataria multiflora using different machine-learning algorithms

Background: Myocardial perfusion scintigraphy (MPS) is an important tool for evaluating ischemia in diabetic populations. However, applications of advanced predictive models like multivariate adaptive regression splines (MARS) to estimate summed stress scores (SSS) are lacking. Methods: In this study, 1028 diabetic women undergoing Thallium-201 MPS were analyzed. The dataset was split into training (80%) and testing (20%) subsets. MARS and multiple linear regression (MLR) models were constructed to predict SSS, and their performance was evaluated using root mean square error (RMSE), relative absolute error (RAE), root relative squared error (RRSE), Mean Absolute Percentage Error (MAPE), and Symmetric Mean Absolute Percentage Error (SMAPE). Results: On the testing dataset, the MARS model outperformed the MLR model across all metrics, with an RMSE of 3.25 compared to 3.89 for MLR, an RAE of 0.52 vs. 0.64, and an RRSE of 0.53 vs. 0.67. Similar trends were observed in MAPE (18.7% vs. 22.1%) and SMAPE (17.3% vs. 20.5%). Conclusions: The superior predictive accuracy of the MARS model suggests its potential to enhance non-invasive myocardial risk stratification in diabetic women.