Partial Dependence Research Articles

Exploring Non-Linear and Synergistic Effects of Street Environment on the Spirit of Place in Historic Districts: Using Multi-Source Data and XGBoost

The fragmented remodeling of historic districts undermines the spirit of place. Understanding the intricate relationship between the neighborhood environment and the spirit of place is essential for sustainable urban development. Current research predominantly relies on case studies and the concept of place, which are subjective and lack specific analysis of how the neighborhood environment shapes the spirit of place. In this study, we examine Chuancheng Street in Handan City as a case study. Utilizing the eXtreme Gradient Boosting (XGBoost) model and multi-source data, combined with SHapley Additive exPlanation (SHAP) and Partial Dependence Plots (PDP), we analyze the non-linear and synergistic effects of the street environment on the spirit of place in historic districts. The findings reveal that (1) the proportion of enduring sociability (PES) on the street significantly shapes the spirit of place, with cultural space elements being less prioritized in perception; (2) PES, green vision index (GVI), Integration_800 m, and mixed land use (MLU) have pronounced non-linear impacts on the spirit of place, with strong threshold effects, and these factors also demonstrate a synergistic effect; (3) There are notable spatial variations in the spirit of place across different blocks, particularly influenced by the authenticity of cultural heritage. This study provides fundamental insights into the spirit of place in historic neighborhoods, enabling a better understanding of complex urban dynamics and informing future street regeneration from a place perspective.

The Spatiotemporal Variability of Soil Available Phosphorus and Potassium in Karst Region: The Crucial Role of Socio-Geographical Factors

The contents of soil available phosphorus (AVP) and potassium (AVK) in karstic mountainous agricultural areas have changed rapidly in recent decades. This temporal variation displays strong spatial heterogeneity due to these areas’ complex topography and anthropogenic activities. Socio-geographical factors can reflect the changes in the natural environment caused by human beings, and our objective is to enhance understanding of their role in explaining the changes of AVP and AVK. In a typical karst region (611.5 km2) with uniform soil parent material and low climatic variability, 255 topsoil samples (138 in 2012 and 117 in 2021) were collected to quantify the temporal AVP and AVK changes. Random forest (RF) and partial dependence plot analyses were conducted to investigate the responses of these changes to socio-geographical factors (distance from the nearest town center [DFT] and village density [VD]), topography, biology, and landscape pattern indexes. The mean values of AVP (48.25 mg kg−1) and AVK (357.67 mg kg−1) in 2021 were significantly (p < 0.01) higher than those in 2012 (28.84 mg kg−1 and 131.67 mg kg−1, respectively). Semi-variance analysis showed strong spatial autocorrelation for AVP and AVK, ranging from 7.29% to 10.95% and 13.31% to 10.33% from 2012 to 2021, respectively. Adding socio-geographical factors can greatly improve the explanatory power of RF modeling for AVP and AVK changes by 19% and 27%, respectively. DFT and VD emerged as the two most important variables affecting these changes, followed by elevation. These three variables all demonstrated clear nonlinear threshold effects on AVP and AVK changes. A strong accumulation of AVP and AVK was observed at DFT < 5 km and VD > 20. The AVP changes increased dramatically when the elevation ranged between 1298 m and 1390 m, while the AVK changes decreased rapidly when the elevation ranged between 1350 m and 1466 m. The interaction effects of DFT and VD with elevation on these changes were also demonstrated. Overall, this study examined the important role of socio-geographical factors and their nonlinear threshold and interaction effects on AVP and AVK changes. The findings help unravel the complex causes of these changes and thus contribute to the design of optimal soil phosphorus and potassium management strategies.

Experimental and data-driven prediction for the impact of free/bond lengths of element specimen on interface characterization of ground anchors

The performance of ground anchor often employed in retaining the stability of onshore structures depends largely on its interface shear behavior. Interface characterization using pullout tests on element specimens of ground anchor is a versatile approach for determining its interface shear behavior. This research combines laboratory experiments and data-driven modeling to investigate the effect of free/bond lengths as a critical configuration condition of element specimen on the interface characterization results of ground anchors. Machine Learning models such as back propagation neural network (BPNN), random forest (RF), and support vector regression (SVR) were employed. Moreover, the particle swarm optimization (PSO) algorithm was used to improve the process. The results indicate that variations in free/bond lengths have a negligible effect on the trends of interface bond curves when bond lengths reach 90–100 mm. Moreover, when the bond and free lengths are set to 80 mm and 20 mm, respectively, the configuration conditions of the element specimen have a limited effect on the average interface shear strength. The PSO-BPNN model provided the most accurate predictions, which closely followed the experimental results compared to RF and SVR models. Compared to the other models, it reached a greater correlation coefficient (0.9975) and a lower root-mean-square error (14.89 kPa). Moreover, Partial dependence plot (PDP) is introduced to visualize the established machine learning model.

Machine learning-based modeling of interface creep behavior of grouted soil anchors with varying soil moistures

The interface creep behavior of the grouted soil anchor subject to varying soil moisture was investigated using the combined incorporation of experimental and data-driven modeling methods to establish an efficient and robust forecasting framework. This study carried out the rapid and creep pullout tests of element anchor specimens at various saturations and then utilized machine learning methods to predict the development of interface creep displacement. The stepwise loading strategy and nonlinear superposition method were combined to generate the interface shear creep curves of the element anchor specimens. A total of 936 data groups of the interface shear displacement were collected with changing soil moisture contents, interface shear time, and interface shearing stress. Next, this study explored the Back Propagation Neural Network (BPNN) and four other machine learning algorithms in predicting the interface creep behavior of the grouted soil anchor under various moisture conditions. As for the hyperparameters, the beetle antennae search (BAS) approach was employed to optimize the BPNN and random forest (RF) models. Finally, the boxplot and Taylor diagrams proved the BAS-BPNN demonstrated a better performance than BAS-RF in predicting the interface creep behavior. The consequent correlation coefficients ranged from 0.9613 to 0.9805 for BPNN, indicating the accuracy and reliability of the interface creep prediction. A partial dependence plot (PDP) was also introduced to visualize the established machine learning model. The threshold of moisture content near 28.7 % is found to switch the interface shear stress-displacement response from strain-stabilizing to strain-softening behavior and to result in the main moisture-increase-induced interface strength degradation. The soil moisture fluctuation leads to the development of interface shear displacement mainly observed in the early phase of 20 h after the onset of moisture change. The uncovered coupled impact of soil moisture condition and interface shear stress state can provide insights into the evaluation of the time-dependent in-service performance of grouted soil anchors embedded in clayey soils.

Development and validation of an automatic machine learning model to predict abnormal increase of transaminase in valproic acid-treated epilepsy.

Valproic acid (VPA) is a primary medication for epilepsy, yet its hepatotoxicity consistently raises concerns among individuals. This study aims to establish an automated machine learning (autoML) model for forecasting the risk of abnormal increase of transaminase levels while undergoing VPA therapy for 1995 epilepsy patients. The study employed the two-tailed T test, Chi-square test, and binary logistic regression analysis, selecting six clinical parameters, including age, stature, leukocyte count, Total Bilirubin, oral dosage of VPA, and VPA concentration. These variables were used to build a risk prediction model using "H2O" autoML platform, achieving the best performance (AUC training = 0.855, AUC test = 0.789) in the training and testing data set. The model also exhibited robust accuracy (AUC valid = 0.742) in an external validation set, underscoring its credibility in anticipating VPA-induced transaminase abnormalities. The significance of the six variables was elucidated through importance ranking, partial dependence, and the TreeSHAP algorithm. This novel model offers enhanced versatility and explicability, rendering it suitable for clinicians seeking to refine parameter adjustments and address imbalanced data sets, thereby bolstering classification precision. To summarize, the personalized prediction model for VPA-treated epilepsy, established with an autoML model, displayed commendable predictive capability, furnishing clinicians with valuable insights for fostering pharmacovigilance.

Ground Reaction Forces and Joint Moments Predict Metabolic Cost in Physical Performance: Harnessing the Power of Artificial Neural Networks

Understanding metabolic cost through biomechanical data, including ground reaction forces (GRFs) and joint moments, is vital for health, sports, and rehabilitation. The long stabilization time (2–5 min) of indirect calorimetry poses challenges in prolonged tests. This study investigated using artificial neural networks (ANNs) to predict metabolic costs from the GRF and joint moment time series. Data from 20 participants collected over 270 walking trials, including the GRF and joint moments, formed a detailed dataset. Two ANN models were crafted, netGRF for the GRF and netMoment for joint moments, and both underwent training, validation, and testing to validate their predictive accuracy for metabolic cost. NetGRF (six hidden layers, two input delays) showed significant correlations: 0.963 (training), 0.927 (validation), 0.883 (testing), p < 0.001. NetMoment (three hidden layers, one input delay) had correlations of 0.920 (training), 0.956 (validation), 0.874 (testing), p < 0.001. The models’ low mean squared errors reflect their precision. Using Partial Dependence Plots, we demonstrated how gait cycle phases affect metabolic cost predictions, pinpointing key phases. Our findings show that the GRF and joint moments data can accurately predict metabolic costs via ANN models, with netGRF being notably consistent. This emphasizes ANNs’ role in biomechanics as a crucial method for estimating metabolic costs, impacting sports science, rehabilitation, assistive technology development, and fostering personalized advancements.

Crash severity prediction and interpretation for road determinants based on a hybrid method

In this study, 4 classification algorithms were employed to characterize the influence of road determinants on roadway crash severity with actual crash data. The crash data were obtained from crash records in Texas, USA, from January 2020 to April 2021. The prediction model of crash severity utilized 12 road-related features—including shoulder types, shoulder width, and curb types—as well as 10 other features—such as weather and illumination conditions—as input features. Three crash severity levels—“Minor Damage,” “Moderate Damage,” and “Severe Damage”—were used as output features. Decision tree, support vector machines, and multi-layer perceptron were employed to compare their prediction performance with the XGBoost model. The results show that the XGBoost model yields the best performance among the 4 algorithms. The overall accuracy, average precision, average recall, and average F1 score of the XGBoost model were 82.65%, 0.83, 0.82, and 0.82, respectively. Besides, SHapley Additive exPlanations (SHAP) and partial dependence plots were used to interpret the model results. Among the road-related features, the most influential one is the median width. Greater crash severity is related to paved right shoulder and curb. These findings are helpful for the design and planning of road safety.

Multi-output neural network model for predicting biochar yield and composition

In biomass pyrolysis for biochar production, existing prediction models face computational challenges and limited accuracy. This study curated a comprehensive dataset, revealing pyrolysis parameters' dominance in biochar yield (54.8 % importance). Pyrolysis temperature emerged as pivotal (PCC = −0.75), influencing yield significantly. Artificial Neural Network (ANN) outperformed Random Forest (RF) in testing set predictions (R2 = 0.95, RMSE = 3.6), making it apt for complex multi-output predictions and software development. The trained ANN model, employed in Partial Dependence Analysis, uncovered nonlinear relationships between biomass characteristics and biochar yield. Findings indicated optimization opportunities, correlating low pyrolysis temperatures, elevated nitrogen content, high fixed carbon, and brief residence times with increased biochar yields. A multi-output ANN model demonstrated optimal fit for biochar yield. A user-friendly Graphical User Interface (GUI) for biochar synthesis prediction was developed, exhibiting robust performance with a mere 0.52 % prediction error for biochar yield. This study showcases practical machine learning application in biochar synthesis, offering valuable insights and predictive tools for optimizing biochar production processes.

Interaction Difference Hypothesis Test for Prediction Models

Machine learning research focuses on the improvement of prediction performance. Progress was made with black-box models that flexibly adapt to the given data. However, due to their increased complexity, black-box models are more difficult to interpret. To address this issue, techniques for interpretable machine learning have been developed, yet there is still a lack of methods to reliably identify interaction effects between predictors under uncertainty. In this work, we present a model-agnostic hypothesis test for the identification of interaction effects in black-box machine learning models. The test statistic is based on the difference between the variance of the estimated prediction function and a version of the estimated prediction function without interaction effects derived via partial dependence functions. The properties of the proposed hypothesis test were explored in simulations of linear and nonlinear models. The proposed hypothesis test can be applied to any black-box prediction model, and the null hypothesis of the test can be flexibly specified according to the research question of interest. Furthermore, the test is computationally fast to apply, as the null distribution does not require the resampling or refitting of black-box prediction models.

Predicting Cd(II) adsorption capacity of biochar materials using typical machine learning models for effective remediation of aquatic environments

The screening and design of “green” biochar materials with high adsorption capacity play a pivotal role in promoting the sustainable treatment of Cd(II)-containing wastewater. In this study, six typical machine learning (ML) models, namely Linear Regression, Random Forest, Gradient Boosting Decision Tree, CatBoost, K-Nearest Neighbors, and Backpropagation Neural Network, were employed to accurately predict the adsorption capacity of Cd(II) onto biochars. A large dataset with 1051 data points was generated using 21 input variables obtained from batch adsorption experiments, including preparation conditions for biochar (2 features), physical properties of biochar (4 features), chemical composition of biochar (9 features), and adsorption experiment conditions (6 features). The rigorous evaluation and comparison of the ML models revealed that the CatBoost model exhibited the highest test R2 value (0.971) and the lowest RMSE (20.54 mg/g), significantly outperforming all other models. The feature importance analysis using Shapley Additive Explanations (SHAP) indicated that biochar chemical compositions had the greatest impact on model predictions of adsorption capacity (42.2 %), followed by adsorption conditions (37.57 %), biochar physical characteristics (12.38 %), and preparation conditions (7.85 %). The optimal experimental conditions optimized by partial dependence plots (PDP) are as follows: as high Cd(II) concentration as possible, C(%) of 33 %, N(%) of 0.3 %, adsorption time of 600 min, pyrolysis time of 50 min, biochar dosage of less than 2 g/L, O(%) of 42 %, biochar pH value of 11.2, and DBE of 1.15. This study unveils novel insights into the adsorption of Cd(II) and provides a comprehensive reference for the sustainable engineering of biochars in Cd(II) wastewater treatment.

Aqueous arsenic (III) removal using a novel solid waste based porous filter media block: Traditional and machine learning (ML) approaches

Arsenic (As) contamination of groundwater is one of the main obstacles preventing people from having easy access to potable water. Gypsum-based a porous filter media block (PFMB) made with moringa bark was proposed to remove arsenic (III) from aqueous solutions. The adsorption isotherms using Langmuir (R2 = 0.976) and Freundlich (R2 = 0.967) and kinetics using pseudo-first and second order and intraparticle diffusion models were achieved. The removal rates of PFMB using 10 g/L in batch experiments were found to be 98.4 % and 94.2 % for 0.50 and 1.0 ppm, respectively. The efficiency was evaluated by varying the material composition, pH, and experimental conditions using a machine learning model with an R2 = 0.95. A new approach to forecasting adsorption capacity based on material composition and experimental data allows the effective determinate to be found from a wider perspective. The model explanation with shapley additive value (SHAP) and partial dependence plots (PDP) showed the initial concentration, and time were the most influential factors in the adsorption process. Arsenic (III) adsorption on the innovative composite material appears promising, which will ensure the reuse and effective disposal of solid waste. ANN can be used to evaluate and predict new conditions without lab experimentation.

Quantitative prediction of postpartum hemorrhage in cesarean section on machine learning

BackgroundCesarean section-induced postpartum hemorrhage (PPH) potentially causes anemia and hypovolemic shock in pregnant women. Hence, it is helpful for obstetricians and anesthesiologists to prepare pre-emptive prevention when predicting PPH occurrence in advance. However, current works on PPH prediction focus on whether PPH occurs rather than assessing PPH amount. To this end, this work studies quantitative PPH prediction with machine learning (ML).MethodsThe study cohort in this paper was selected from individuals with PPH who were hospitalized at Shijiazhuang Obstetrics and Gynecology Hospital from 2020 to 2022. In this study cohort, we built a dataset with 6,144 subjects covering clinical parameters, anesthesia operation records, laboratory examination results, and other information in the electronic medical record system. Based on our built dataset, we exploit six different ML models, including logistic regression, linear regression, gradient boosting, XGBoost, multilayer perceptron, and random forest, to automatically predict the amount of bleeding during cesarean section. Eighty percent of the dataset was used as model training, and 20%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} was used for verification. Those ML models are constantly verified and improved by root mean squared error(RMSE) and mean absolute error(MAE). Moreover, we also leverage the importance of permutation and partial dependence plot (PDP) to discuss their feasibility.ResultThe experiment results show that random forest obtains the highest accuracy for PPH amount prediction compared to other ML methods. Random forest reaches the mean absolute error of 21.7, less than 5.4%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} prediction error. It also gains the root mean squared error of 33.75, less than 9.3%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} prediction error. On the other hand, the experimental results also disclose indicators that contributed most to PPH prediction, including Ca, hemoglobin, white blood cells, platelets, Na, and K.ConclusionIt effectively predicts the amount of PPH during a cesarean section by ML methods, especially random forest. With the above insight, ML predicting PPH amounts provides early warning for clinicians, thus reducing complications and improving cesarean sections’ safety. Furthermore, the importance of ML and permutation, complemented by incorporating PDP, promises to provide clinicians with a transparent indication of individual risk prediction.

IPB-MSA&SO4: a daily 0.25° resolution dataset of in situ-produced biogenic methanesulfonic acid and sulfate over the North Atlantic during 1998–2022 based on machine learning

Abstract. Accurate long-term marine-derived biogenic sulfur aerosol concentrations at high spatial and temporal resolutions are critical for a wide range of studies, including climatology, trend analysis, and model evaluation; this information is also imperative for the accurate investigation of the contribution of marine-derived biogenic sulfur aerosol concentrations to the aerosol burden, for the elucidation of their radiative impacts, and to provide boundary conditions for regional models. By applying machine learning algorithms, we constructed the first publicly available daily gridded dataset of in situ-produced biogenic methanesulfonic acid (MSA) and non-sea-salt sulfate (nss-SO4=) concentrations covering the North Atlantic. The dataset is of high spatial resolution (0.25° × 0.25°) and spans 25 years (1998–2022), far exceeding what observations alone could achieve both spatially and temporally. The machine learning models were generated by combining in situ observations of sulfur aerosol data from Mace Head Atmospheric Research Station, located on the west coast of Ireland, and from the North Atlantic Aerosols and Marine Ecosystems Study (NAAMES) cruises in the northwestern Atlantic with the constructed sea-to-air dimethylsulfide flux (FDMS) and ECMWF ERA5 reanalysis datasets. To determine the optimal method for regression, we employed five machine learning model types: support vector machines, decision tree, regression ensemble, Gaussian process regression, and artificial neural networks. A comparison of the mean absolute error (MAE), root-mean-square error (RMSE), and coefficient of determination (R2) revealed that Gaussian process regression (GPR) was the most effective algorithm, outperforming the other models with respect to simulating the biogenic MSA and nss-SO4= concentrations. For predicting daily MSA (nss-SO4=), GPR displayed the highest R2 value of 0.86 (0.72) and the lowest MAE of 0.014 (0.10) µg m−3. GPR partial dependence analysis suggests that the relationships between predictors and MSA and nss-SO4= concentrations are complex rather than linear. Using the GPR algorithm, we produced a high-resolution daily dataset of in situ-produced biogenic MSA and nss-SO4= sea-level concentrations over the North Atlantic, which we named “In-situ Produced Biogenic Methanesulfonic Acid and Sulfate over the North Atlantic” (IPB-MSA&SO4). The obtained IPB-MSA&SO4 data allowed us to analyze the spatiotemporal patterns of MSA and nss-SO4= as well as the ratio between them (MSA:nss-SO4=). A comparison with the existing Copernicus Atmosphere Monitoring Service ECMWF Atmospheric Composition Reanalysis 4 (CAMS-EAC4) reanalysis suggested that our high-resolution dataset reproduces the spatial and temporal patterns of the biogenic sulfur aerosol concentration with high accuracy and has high consistency with independent measurements in the Atlantic Ocean. IPB-MSA&SO4 is publicly available at https://doi.org/10.17632/j8bzd5dvpx.1 (Mansour et al., 2023b).

Exploring the variation in associations between socioeconomic indicators and non-communicable diseases in the Tromsø Study: an algorithmic approach

Aims: We contribute to the methodological literature on the assessment of health inequalities by applying an algorithmic approach to evaluate the capabilities of socioeconomic variables in predicting the prevalence of non-communicable diseases in a Norwegian health survey. Methods: We use data from the seventh survey of the population based Tromsø Study (2015–2016), including 11,074 women and 10,009 men aged 40 years and above. We apply the random forest algorithm to predict four non-communicable disease outcomes (heart attack, cancer, diabetes and stroke) based on information on a number of social root causes and health behaviours. We evaluate our results using the classification error, the mean decrease in accuracy, partial dependence statistics. Results: Results suggest that education, household income and occupation to a variable extent contribute to predicting non-communicable disease outcomes. Prediction misclassification ranges between 25.1% and 35.4% depending on the non-communicable diseases under study. Partial dependences reveal mostly expected health gradients, with some examples of complex functional relationships. Out-of-sample model validation shows that predictions translate to new data input. Conclusions: Algorithmic modelling can provide additional empirical detail and metrics for evaluating heterogeneous inequalities in morbidity. The extent to which education, income and occupation contribute to predicting binary non-communicable disease outcomes depends on both non-communicable diseases and socioeconomic indicator. Partial dependences reveal that social gradients in non-communicable disease outcomes vary in shape between combinations of non-communicable disease outcome and socioeconomic status indicator. Misclassification rates highlight the extent of variation within socioeconomic groups, suggesting that future studies may improve predictive accuracy by exploring further subpopulation heterogeneity.

Modifiable factors associated with Huntington's disease progression in presymptomatic participants.

Huntington's disease (HD) is a neurodegenerative disorder characterized by progressive motor, cognitive, and psychiatric symptoms. Our aim here was to identify factors that can be modified to slow disease progression even before the first symptoms appear. We included 2636 presymptomatic individuals (comparison with family controls) drawn from the prospective observational cohort Enroll-HD, with more than 35 CAG repeats and at least two assessments of disease progression measured with the composite Huntington's disease rating Scale (cUHDRS). The association between sociodemographic factors, health behaviors, health history, and cUHDRS trajectory was assessed with a mixed-effects random forest using partial dependence plots and Shapley additive explanation method. Participants were followed by an average of 3.4 (SD = 1.97) years. We confirmed the negative impact of age and a high number of CAG repeats. We found that a high level of education, a body mass index (BMI) <23 kg/m2 before the age of 40 and >23 kg/m2 thereafter, alcohol consumption of <15 units per week, current coffee consumption and no smoking were linked to slow disease progression, as did no previous exposure to antidepressants or anxiolytic, no psychiatric history or comorbidities, and being female. Other comorbidities or marital status showed no major association with HD evolution. Reducing modifiable risk factors for HD is one way to support the presymptomatic population. A high level of education, low-to-moderate alcohol consumption, no smoking, and BMI control are likely to slow disease progression in this population.

Machine learning for predicting compressive strength of sustainable cement paste incorporating copper mine tailings as supplementary cementitious materials

Copper mining produces significant amounts of copper mine tailings (CMT), necessitating appropriate waste handling and disposal practices. By substituting a portion of cement with CMT as supplementary cementitious materials (SCMs), we aim to address two environmental issues simultaneously: reducing copper mine waste in landfills and decreasing embodied carbon by using less cement. The exploration of CMT recycling as a cement replacement requires evaluation of its impact on material performance, such as compressive strength. In this paper, we address this by machine learning that features data fusion of large public data with our own small data on compressive strength of CMT-incorporated cement. We developed and critically evaluated three machine learning models: a simple linear model, Gaussian process, and random forest that predict the compressive strength of CMT-incorporated cement pastes with different mix designs (e.g., varying amounts of CMT and water-binder ratios) and curing ages. Hyperparameters in the random forest model were tuned using Bayesian optimization. Following a comprehensive evaluation of the models, we find that the random forest model can accurately estimate the compressive strength of cement paste across the mix designs. Furthermore, results from SHapley Additive exPlanation (SHAP), Individual Conditional Expectation (ICE), and Partial Dependence Plots (PDP) revealed that cement, ground-granulated blast furnace slag, superplasticizers, and curing ages positively influence compressive strength. This study contributes to acceleration of sustainable material technology to obtain the best mix design and desired compressive strength.

Exploring the Effectiveness of SHAP over other Explainable AI Methods

Explainable Artificial Intelligence (XAI) has emerged as a critical domain to demystify the opaque decision-making processes of machine learning models, fostering trust and understanding among users. Among various XAI methods, SHAP (SHapley Additive exPlanations) has gained prominence for its theo- retically grounded approach and practical applicability. The paper presents a comprehensive exploration of SHAP’s effectiveness compared to other promi- nent XAI methods.Methods such as LIME (Local Interpretable Model-agnostic Explanations), permutation importance, Anchors and partial dependence plots are examined for their respective strengths and limitations. Through a detailed analysis of their principles, strengths, and limitations through reviewing differ- ent research papers based on some important factors of XAI, the paper aims to provide insights into the effectiveness and suitability of these methods.The study offers valuable guidance for researchers and practitioners seeking to incorporate XAI into their AI systems. Keywords: SHAP, XAI, LIME, permutation importance, Anchors and par- tial dependence plots

Predicting the temporal-spatial distribution of chlorophyll-a in the Yellow River estuary using explainable machine learning

The water quality in estuaries is influenced by both physical and biochemical processes, making it difficult to identify and quantify the main controlling factors of chlorophyll-a (Chl-a) concentration. This study introduces an interpretable machine learning approach to identify and quantify the dominant control factors determining Chl-a concentration in the Yellow River Estuary (YRE). The model utilizes in situ data and MODIS data in the Yellow River Estuary, incorporating five key variables: salinity, water depth, turbidity, dissolved inorganic nitrogen (DIN), and soluble reactive phosphorus (SRP). Three types of inputs are used to predict Chl-a, including physical, biogeochemical, and physical-biogeochemical factors. The model further uses feature importance ranking and partial dependence plot to identify and quantify key control factors. The results reveal that (1) Chl-a variation is influenced by multiple factors, especially salinity and turbidity; (2) Chl-a concentration is inhibited when salinity is below 26, turbidity is lower than 16 NTU and above 30 NTU, DIN is lower than 0.314 mg/L and SRP is below 0.01 mg/L; (3) riverine inputs change turbidity and salinity in the Yellow River Estuary and Laizhou Bay, in response to temporal-spatial distributions of phytoplankton. This research can provide a valuable reference for water quality management in the Yellow River Estuary.

Explainable machine learning predicts survival of retroperitoneal liposarcoma: A study based on the SEER database and external validation in China

AbstractObjectiveWe have developed explainable machine learning models to predict the overall survival (OS) of retroperitoneal liposarcoma (RLPS) patients. This approach aims to enhance the explainability and transparency of our modeling results.MethodsWe collected clinicopathological information of RLPS patients from The Surveillance, Epidemiology, and End Results (SEER) database and allocated them into training and validation sets with a 7:3 ratio. Simultaneously, we obtained an external validation cohort from The First Affiliated Hospital of Naval Medical University (Shanghai, China). We performed LASSO regression and multivariate Cox proportional hazards analysis to identify relevant risk factors, which were then combined to develop six machine learning (ML) models: Cox proportional hazards model (Coxph), random survival forest (RSF), ranger, gradient boosting with component‐wise linear models (GBM), decision trees, and boosting trees. The predictive performance of these ML models was evaluated using the concordance index (C‐index), the integrated cumulative/dynamic area under the curve (AUC), and the integrated Brier score, as well as the Cox–Snell residual plot. We also used time‐dependent variable importance, analysis of partial dependence survival plots, and the generation of aggregated survival SHapley Additive exPlanations (SurvSHAP) plots to provide a global explanation of the optimal model. Additionally, SurvSHAP (t) and survival local interpretable model‐agnostic explanations (SurvLIME) plots were used to provide a local explanation of the optimal model.ResultsThe final ML models are consisted of six factors: patient's age, gender, marital status, surgical history, as well as tumor's histopathological classification, histological grade, and SEER stage. Our prognostic model exhibits significant discriminative ability, particularly with the ranger model performing optimally. In the training set, validation set, and external validation set, the AUC for 1, 3, and 5 year OS are all above 0.83, and the integrated Brier scores are consistently below 0.15. The explainability analysis of the ranger model also indicates that histological grade, histopathological classification, and age are the most influential factors in predicting OS.ConclusionsThe ranger ML prognostic model exhibits optimal performance and can be utilized to predict the OS of RLPS patients, offering valuable and crucial references for clinical physicians to make informed decisions in advance.

Lower limb strength and associated factors among older patients with dementia

This cross-sectional study was conducted to assess lower limb strength and related factors among older people with dementia examined and treated at the National Geriatric Hospital. Lower limb strength was assessed using the 30-Second Chair Stand Test. The proportion of participants with impaired lower limb strength was 78.2%. There were significant associations between impaired lower limb strength and diabetic patients. There were significant associations between impaired lower limb strength and geriatric syndromes: malnutrition, high risk of falls, and dependence or partial dependence according to the Barthel Index. There was no significant difference between the type of dementia and the level of dementia with impaired lower limb strength. In conclusion: Three-quarters of older dementia patients have impaired lower limb strength. Our results highlighted that diabetes, and geriatric syndromes such as malnutrition, high fall risk, dependence, and partial dependence were significantly associated with lower limb strength in older patients with dementia.