Partial Dependence Research Articles

Оперативное измерение горной массы в процессе погрузочно-доставочных работ

The article discusses an automated control system for underground mining of copper-nickel ores and the weight control system as a part of this system. The main functions of the weight control system are described, as well as the requirements towards this system, the principle of applying the proposed algorithms and the results of implementing this system at an existing mining operation. The experience of measuring the rock mass during loading and hauling operations is described in terms of operational management and control of actual quantitative and qualitative parameters of ore flow in the multistage process of the ore grade stabilization in underground mining. The obtained results help to solve the challenge of controlling the LHD bucket load while ensuring the required parameters and operating conditions for the following application scenarios: (1) control of the transported rock mass for each haul cycle with the accuracy of ±6%; (2) control of the transported rock mass by a particular LHD per shift with the accuracy not worse than 1% for 60 haul cycles per shift; (3) for the operations with high moisture content of the transported load, it is possible to account for the stuck material for each haul cycle; (4) no special requirements to the weighing process; (5) minimization of the operating costs for calibration and maintenance; (6) high reliability due to minimum amount of hardware; (7) no degradation of the method when using readings of the hydraulic system; (8) unification of the system. The accuracy and reliability of the data obtained gradually increases during the operation when the test measurements are made about once a month. The disadvantages include: (1) impossibility to use the metrological level for measurements; (2) dependence on the technical condition of the onboard hydraulic system; (3) partial dependence on the LHD mileage. A conclusion is made that the process of ore grade stabilization should be multistage, and the measurement of the rock mass during the loading and hauling operations is one of the mandatory methods for implementation of the geometallurgical tasks.

Quantifying the Impact of Environment Loads on Displacements in a Suspension Bridge with a Data-Driven Approach.

Long-span bridges are susceptible to damage, aging, and deformation in harsh environments for a long time. Therefore, structural health monitoring (SHM) systems need to be used for reasonable monitoring and maintenance. Among various indicators, bridge displacement is a crucial parameter reflecting the bridge's health condition. Due to the simultaneous bearing of multiple environmental loads on suspension bridges, determining the impact of different loads on displacement is beneficial for the better understanding of the health conditions of the bridges. Considering the fact that extreme gradient boosting (XGBoost) has higher prediction performance and robustness, the authors of this paper have developed a data-driven approach based on the XGBoost model to quantify the impact between different environmental loads and the displacement of a suspension bridge. Simultaneously, this study combined wavelet threshold (WT) denoising and the variational mode decomposition (VMD) method to conduct a modal decomposition of three-dimensional (3D) displacement, further investigating the interrelationships between different loads and bridge displacements. This model links wind speed, temperature, air pressure, and humidity with the 3D displacement response of the span using the bridge monitoring data provided by the GNSS and Earth Observation for Structural Health Monitoring (GeoSHM) system of the Forth Road Bridge (FRB) in the United Kingdom (UK), thus eliminating the temperature time-lag effect on displacement data. The effects of the different loads on the displacement are quantified individually with partial dependence plots (PDPs). Employing testing, it was found that the XGBoost model has a high predictive effect on the target variable of displacement. The analysis of quantification and correlation reveals that lateral displacement is primarily affected by same-direction wind, showing a clear positive correlation, and vertical displacement is mainly influenced by temperature and exhibits a negative correlation. Longitudinal displacement is jointly influenced by various environmental loads, showing a positive correlation with atmospheric pressure, temperature, and vertical wind and a negative correlation with longitudinal wind, lateral wind, and humidity. The results can guide bridge structural health monitoring in extreme weather to avoid accidents.

Machine learning-based deoxidizer screening for intensified hydrogen production from steam splitting

The design of advanced deoxidizer is the key to promote hydrogen production from chemical looping steam splitting, however, the deoxidizer shows complicated possibility of composition, which results in long duration in material exploitation. In this study, Gibbs free energy change (ΔG) is used as the output of the model, and three machine learning models, Decision Tree, Random Forest, and Gradient Boosting Tree algorithms, are established and optimized for functionalized deoxidizer screening. Results indicate that the Gradient Boosting Tree algorithm shows the best performance, and the RMSE and R2 of the test dataset reach 22.24 and 0.93. Through the analysis of feature importance and Partial Dependence Plots, the effects of different deoxidizer properties on the difficulty of steam splitting can be intuitively displayed. Specifically, 907 groups of feasible deoxidizers were selected from 772,083 ones through the combination of machine learning and high-throughput screening. Steam splitting experiments demonstrate the prediction accuracy of 81.3% by Gradient Boosting Tree model. Several less reported composite deoxidizers such as Y2O3+Fe, SrO + Fe, SrO + Co, Li2O + Fe, Li2O + Co were predicted and experimentally verified. This study provides prominent strategy and new implications for deoxidizer screening, which will substantially promote the sustainable development of steam splitting hydrogen generation.

Peptidoglycan synthesis drives a single population of septal cell wall synthases during division in Bacillus subtilis

Bacterial cell division requires septal peptidoglycan (sPG) synthesis by the divisome complex. Treadmilling of the essential tubulin homologue FtsZ has been implicated in septal constriction, though its precise role remains unclear. Here we used live-cell single-molecule imaging of the divisome transpeptidase PBP2B to investigate sPG synthesis dynamics in Bacillus subtilis. In contrast to previous models, we observed a single population of processively moving PBP2B molecules whose motion is driven by peptidoglycan synthesis and is not associated with FtsZ treadmilling. However, despite the asynchronous motions of PBP2B and FtsZ, a partial dependence of PBP2B processivity on FtsZ treadmilling was observed. Additionally, through single-molecule counting experiments we provide evidence that the divisome synthesis complex is multimeric. Our results support a model for B. subtilis division where a multimeric synthesis complex follows a single track dependent on sPG synthesis whose activity and dynamics are asynchronous with FtsZ treadmilling.

Investigation of light gas adsorption by microporous materials using data analysis and decision tree machine learning algorithm

This study aimed to extract insights into the adsorption behavior of light gases by microporous materials such as zeolites, MOFs, and activated carbons. Data reported in 22 articles published between 1974 and 2022 were analyzed using a decision tree (DT) machine learning algorithm. A comprehensive database comprising 3297 data points, elucidating the impacts of 8 input variables on adsorption capacity, was constructed. Various exploratory data analysis techniques, including histograms, bar charts, and scatter plots, were employed to discern the relationships among input variables and the performance variable. Additionally, a DT model was employed to regress the adsorbed capacity data. Furthermore, a parametric study of this model facilitated the determination of the relative importance of input variables and their partial dependence, enhancing the interpretability of the model and enabling the deduction of heuristics for high or low adsorption capacity. The exploratory data analysis revealed that pressure and molecular mass of the adsorbate were the most significant variables influencing adsorption capacity.

Ascertaining sensitive exposure biomarkers of various metal(loid)s to embryo implantation

Close relationships exist between metal(loid)s exposure and embryo implantation failure (EIF) from animal and epidemiological studies. However, there are still inconsistent results and lacking of sensitive metal(loid) exposure biomarkers associated with EIF risk. We aimed to ascertain sensitive metal(loid) biomarkers to EIF and provide potential biological explanations. Candidate metal(loid) biomarkers were measured in the female hair (FH), female serum (FS), and follicular fluid (FF) with various exposure time periods. An analytical framework was established by integrating epidemiological association results, comprehensive literature searching, and knowledge-based adverse outcome pathway (AOP) networks. The sensitive biomarkers of metal(loid)s along with potential biological pathways to EIF were identified in this framework. Among the concerned 272 candidates, 45 metal(loid)s biomarkers across six time periods and three biomatrix were initially identified by single-metal(loid) analyses. Two biomarkers with counterfactual results according to literature summary results were excluded, and a total of five biomarkers were further determined from 43 remained candidates in mixture models. Finally, four sensitive metal(loid) biomarkers were eventually assessed by overlapping AOP networks information, including Se and Co in FH, and Fe and Zn in FS. AOP networks also identified key GO pathways and proteins involved in regulation of oxygen species biosynthetic, cell proliferation, and inflammatory response. Partial dependence results revealed Fe in FS and Co in FH at their low levels might be potential sensitive exposure levels for EIF. Our study provided a typical framework to screen the crucial metal(loid) biomarkers and ascertain that Se and Co in FH, and Fe and Zn in FS played an important role in embryo implantation.

Predicting Redox Conditions in Groundwater at a National Scale Using Random Forest Classification.

Redox conditions in groundwater may markedly affect the fate and transport of nutrients, volatile organic compounds, and trace metals, with significant implications for human health. While many local assessments of redox conditions have been made, the spatial variability of redox reaction rates makes the determination of redox conditions at regional or national scales problematic. In this study, redox conditions in groundwater were predicted for the contiguous United States using random forest classification by relating measured water quality data from over 30,000 wells to natural and anthropogenic factors. The model correctly predicted the oxic/suboxic classification for 78 and 79% of the samples in the out-of-bag and hold-out data sets, respectively. Variables describing geology, hydrology, soil properties, and hydrologic position were among the most important factors affecting the likelihood of oxic conditions in groundwater. Important model variables tended to relate to aquifer recharge, groundwater travel time, or prevalence of electron donors, which are key drivers of redox conditions in groundwater. Partial dependence plots suggested that the likelihood of oxic conditions in groundwater decreased sharply as streams were approached and gradually as the depth below the water table increased. The probability of oxic groundwater increased as base flow index values increased, likely due to the prevalence of well-drained soils and geologic materials in high base flow index areas. The likelihood of oxic conditions increased as topographic wetness index (TWI) values decreased. High topographic wetness index values occur in areas with a propensity for standing water and overland flow, conditions that limit the delivery of dissolved oxygen to groundwater by recharge; higher TWI values also tend to occur in discharge areas, which may contain groundwater with long travel times. A second model was developed to predict the probability of elevated manganese (Mn) concentrations in groundwater (i.e., ≥50 μg/L). The Mn model relied on many of the same variables as the oxic/suboxic model and may be used to identify areas where Mn-reducing conditions occur and where there is an increased risk to domestic water supplies due to high Mn concentrations. Model predictions of redox conditions in groundwater produced in this study may help identify regions of the country with elevated groundwater vulnerability and stream vulnerability to groundwater-derived contaminants.

Machine learning-driven innovations in green eco-environmental rubberized concrete design towards sustainability

This research investigates the state-of-the-art developments in environmentally sustainable rubberized concrete (RC) design, with a significant focus on its eco-friendly characteristics. Novel methodologies for enhancing RC composition are currently under investigation, and machine learning, specifically CatBoost, plays a pivotal role in facilitating this exploration. The model is employed on a dataset comprising 590 RC samples. Importantly, an impressive coefficient of determination value of 0.991 is attained while using CatBoost to predict the compressive strength (CS) of RC, highlighting its exceptional predictive accuracy. Visual representations of variable interactions are generated using Partial Dependence Plots (PDPs). This study further presents design optimization results and key insights, paving the way for the development of sustainable RC materials. In addition, Pareto optimal solutions are presented, prioritizing the maximization of rubber and cementitious replacement material content while simultaneously striving for optimal CS. This approach embodies the study’s dedication to achieving a balanced and sustainable outcome where environmental benefits are maximized without compromising material performance.

Temporal variables improve a spatiotemporal species distribution model for the non-native freshwater fish Candidia temminckii

Ecosystem conservation requires a deeper understanding of species-habitat relationships and population dynamics at a fine spatiotemporal resolution. We propose a new distribution modeling method based on a 5-year monthly survey that considers the temporal continuity of species distributions and physical habitat datasets by inputting continuous time-related variables. We employed random forests to relate the presence/absence of the non-native freshwater fish Candidia temminckii to physical habitat data at 15 sampling sites along a 1.4km spring-fed river in Japan. The proposed method outperforms all conventional methods using datasets split into a specific time period to incorporate temporality into the model. The order of variable importance and shape of the partial dependence plots of the proposed method reflect species ecology and show a gradual shift over time compared to the conventional methods. These results demonstrate the applicability of the proposed method to species distribution modeling using fine-scale spatiotemporal data.

Applying explainable artificial intelligence methods to models for diagnosing personal traits and cognitive abilities by social network data

This study utilizes advanced artificial intelligence techniques to analyze the social media behavior of 1358 users on VK, the largest Russian online social networking service. The analysis comprises 753,252 posts and reposts, combined with Big Five personality traits test results, as well as assessments of verbal and fluid intelligence. The objective of this research is to understand the manifestation of psychological attributes in social media users' behavior and determine their implications on user-interaction models. We employ the integrated gradients method to identify the most influential feature groups. The partial dependence plot technique aids in understanding how these features function across varying severity degrees of the predicted trait. To evaluate feature stability within the models, we cluster calculated Shapley values. Our findings suggest that the emotional tone (joy, surprise, anger, fear) of posts significantly influences the prediction of three personality traits: Extraversion, Agreeableness, and Openness to Experience. Additionally, user social engagement metrics (such as friend count, subscribers, likes, views, and comments) correlate directly with the predicted level of Logical thinking. We also observe a trend towards provocative and socially reprehensible content among users with high Neuroticism levels. The theme of religion demonstrates a multidirectional relationship with Consciousness and Agreeableness. Further findings, including an analysis of post frequency and key text characteristics, are also discussed, contributing to our understanding of the complex interplay between social media behavior and psychological traits. The study proposes a transition from the analysis of correlations between psychological (cognitive) traits to the analysis of indicators of behavior in a social network that are significant for diagnostic models of the corresponding traits.

Enhancing biomass Pyrolysis: Predictive insights from process simulation integrated with interpretable Machine learning models

Waste biomass pyrolysis is a promising thermochemical conversion process for the production of biofuels and sustainable materials. However, it is challenging to accurately predict the properties and yield of products formed during pyrolysis. Machine learning (ML) is a useful tool for predicting the performance of a process. In the present study, ML algorithms integrated with process simulation were explored to accurately model waste biomass pyrolysis based on properties such as H/C, O/C, oil yield, gas yield, and char yield. Six different ML models including Random Forest (RF), Gradient Boosting Regressor (GBR), eXtreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Artificial Neural Network (ANN), and Stochastic Gradient Descent (SGD) were used to model pyrolysis process. It was found that the out-of-the-box (without optimization) models for RF, XGBoost, ANN, and GBR performed the best and did not benefit from hyperparameter optimization. The GBR was identified as the most effective among various ML models. It accurately predicted yields of gas, biochar, bio-oil yields, and their H/C and O/C compositions. GBR effectively demonstrated the complex relationships between these variables. The box plot showing the root mean squared logarithmic error (RMSE) revealed that the GBR model had the best overall performance with a value less than 0.03. Also, the partial dependence plot and SHAP feature importance were evaluated to better understand each feature's effect on the output. Lastly, a shareable graphical user interface (GUI) was created to enable researchers explore and predict pyrolysis yield.

Interpretable and Intuitive Machine Learning Approaches for Predicting Disability Progression in Relapsing-Remitting Multiple Sclerosis Based on Clinical and Gray Matter Atrophy Indicators

Rationale and objectivesTo investigate whether clinical and grey matter (GM) atrophy indicators can predict disability in relapsing-remitting multiple sclerosis (RRMS) and to enhance the interpretability and intuitiveness of a predictive machine learning model. Materials and methodsOne hundred forty-five and 50 RRMS patients with structural MRI and at least 1-year follow-up Expanded Disability Status Scale (EDSS) results were retrospectively enrolled and placed in the discovery and external test cohorts, respectively. Six clinical and radiomics feature-based machine learning classifiers were trained and tested to predict disability progression in the discovery cohort and validated in the external test set. Partial dependence plot (PDP) analysis and a Shiny web application were conducted to enhance the interpretability and intuitiveness. ResultsIn the discovery cohort, 98 patients had disability stability, and 47 patients were classified as having disability progression. In the external test set, 35 patients were disability stable, and 15 patients had disability progression. Models trained with both clinical and radiomics features (area under the curve (AUC), 0.725-0.950) outperformed those trained with clinical (AUC, 0.600-0.740) or radiomics features only (AUC, 0.615-0.945). Among clinical + radiomics feature models, the logistic regression (LR) classifier-based model performed best, with an AUC of 0.950. Only the radiomics feature-only models were applied in the external test set due to the data collection problem and showed fair performance, with AUCs ranging from 0.617 to 0.753. PDP analysis showed that female patients and those with lower volume, surface area, and symbol digit modalities test (SDMT) scores; greater mean curvature and age; and no disease modifying therapy (DMT) had increased probabilities of disease progression. Finally, a Shiny web application (https://lauralin1104.shinyapps.io/LRshiny/) was developed to calculate the risk of disability progression. ConclusionsInterpretable and intuitive machine learning approaches based on clinical and GM atrophy indicators can help physicians predict disability progression in RRMS patients for clinical decision-making and patient management.

Phase Transformation Temperature Prediction in Steels via Machine Learning.

The phase transformation temperature plays an important role in the design, production and heat treatment process of steels. In the present work, an improved version of the gradient-boosting method LightGBM has been utilized to study the influencing factors of the four phase transformation temperatures, namely Ac1, Ac3, the martensite transformation start (MS) temperature and the bainitic transformation start (BS) temperature. The effects of the alloying element were discussed in detail by comparing their influencing mechanisms on different phase transformation temperatures. The training accuracy was significantly improved by further introducing appropriate features related to atomic parameters. The melting temperature and coefficient of linear thermal expansion of the pure metals corresponding to the alloying elements, atomic Waber-Cromer pseudopotential radii and valence electron number were the top four among the eighteen atomic parameters used to improve the trained model performance. The training and prediction processes were analyzed using a partial dependence plot (PDP) and Shapley additive explanation (SHAP) methods to reveal the relationships between the features and phase transformation temperature.

Predicting and analysing initiator crime environments based on machine learning for improving urban safety

PurposeInitiator crimes, also known as near-repeat crimes, occur in places with known risk factors and vulnerabilities based on prior crime-related experiences or information. Consequently, the environment in which initiator crimes occur might be different from more general crime environments. This study aimed to analyse the differences between the environments of initiator crimes and general crimes, confirming the need for predicting initiator crimes.Design/methodology/approachWe compared predictive models using data corresponding to initiator crimes and all residential burglaries without considering repetitive crime patterns as dependent variables. Using random forest and gradient boosting, representative ensemble models and predictive models were compared utilising various environmental factor data. Subsequently, we evaluated the performance of each predictive model to derive feature importance and partial dependence based on a highly predictive model.FindingsBy analysing environmental factors affecting overall residential burglary and initiator crimes, we observed notable differences in high-importance variables. Further analysis of the partial dependence of total residential burglary and initiator crimes based on these variables revealed distinct impacts on each crime. Moreover, initiator crimes took place in environments consistent with well-known theories in the field of environmental criminology.Originality/valueOur findings indicate the possibility that results that do not appear through the existing theft crime prediction method will be identified in the initiator crime prediction model. Emphasising the importance of investigating the environments in which initiator crimes occur, this study underscores the potential of artificial intelligence (AI)-based approaches in creating a safe urban environment. By effectively preventing potential crimes, AI-driven prediction of initiator crimes can significantly contribute to enhancing urban safety.

Effects of Various Heavy Metal Exposures on Insulin Resistance in Non-diabetic Populations: Interpretability Analysis from Machine Learning Modeling Perspective.

Increasing and compelling evidence has been proved that heavy metal exposure is involved in the development of insulin resistance (IR). We trained an interpretable predictive machine learning (ML) model for IR in the non-diabetic populations based on levels of heavy metal exposure. A total of 4354 participants from the NHANES (2003-2020) with complete information were randomly divided into a training set and a test set. Twelve ML algorithms, including random forest (RF), XGBoost (XGB), logistic regression (LR), GaussianNB (GNB), ridge regression (RR), support vector machine (SVM), multilayer perceptron (MLP), decision tree (DT), AdaBoost (AB), Gradient Boosting Decision Tree (GBDT), Voting Classifier (VC), and K-Nearest Neighbour (KNN), were constructed for IR prediction using the training set. Among these models, the RF algorithm had the best predictive performance, showing an accuracy of 80.14%, an AUC of 0.856, and an F1 score of 0.74 in the test set. We embedded three interpretable methods, the permutation feature importance analysis, partial dependence plot (PDP), and Shapley additive explanations (SHAP) in RF model for model interpretation. Urinary Ba, urinary Mo, blood Pb, and blood Cd levels were identified as the main influencers of IR. Within a specific range, urinary Ba (0.56-3.56µg/L) and urinary Mo (1.06-20.25µg/L) levels exhibited the most pronounced upwards trend with the risk of IR, while blood Pb (0.05-2.81µg/dL) and blood Cd (0.24-0.65µg/L) levels showed a declining trend with IR. The findings on the synergistic effects demonstrated that controlling urinary Ba levels might be more crucial for the management of IR. The SHAP decision plot offered personalized care for IR based on heavy metal control. In conclusion, by utilizing interpretable ML approaches, we emphasize the predictive value of heavy metals for IR, especially Ba, Mo, Pb, and Cd.

Exploring the injury severity of vulnerable road users to truck crashes by ensemble learning

Large numbers of vulnerable road users were killed in truck crashes. In this study, ensemble machine learning models are constructed to predict the injury severity of the vulnerable road user (VRU) to truck (VRU-T) crashes. The study is based on the five years (2017–2021) of VRU-T crash data in the Shandong Province from the Center for Accident Research in Zibo. The injury severity of VRUs is estimated using machine learning ensemble models- Stacking, Voting, Random Forest, and eXtreme Gradient Boosting (XGBoost). Compared to the other three models, the Stacking has excellent predictive performance on the pedestrian and non-motorized datasets. Then, SHapley Additive exPlanations and Partial Dependence Plot box are introduced to analyze risk factors qualitatively and quantitatively. The innovative findings of this study are as follows: (1) as VRUs age, they are more likely to be seriously injured in truck crashes; (2) middle-aged truck drivers and truck drivers with medium driving experience increase the probability of VRUs being severe and fatally injured in truck crashes; (3) crashes involving heavy trucks, under signalized crossing, or on the national and provincial road and urban road have a positive effect on the crash severity for cyclists, and E-Bike riders.

The use of machine learning techniques to investigate the properties of metakaolin-based geopolymer concrete

The construction industry significantly contributes to global greenhouse gas emissions, highlighting the imperative for developing environmentally friendly construction materials. Geopolymers, particularly those utilizing metakaolin (MK), have emerged as a promising green alternative to conventional concrete. However, the acquisition of MK-based geopolymer concrete with optimal mechanical properties poses challenges due to numerous influential factors, disagreement over various findings, and the lack of a reliable predictive model. This study aimed to address this gap by employing a wide range of machine learning methods, namely gradient boosting machine, random forest, decision tree, artificial neural network, and support vector machine. Different optimization and regularization techniques were used to comprehensively understand the factors affecting the compressive strength of MK-based geopolymer concrete, including mixture design, chemical characteristics of the initial binder and activators, and different curing regimes. The results demonstrated the exceptional performance of the gradient boosting machine in predicting the compressive strength of MK-based geopolymer concrete, achieving a coefficient of determination of 0.983 and a mean absolute error of 1.615 MPa. Additionally, the study employed partial dependence plots, feature importance analysis, and SHapley Additive exPlanations (SHAP) to elucidate the proposed models. The coarse-to-fine aggregate ratio, H2O/Na2O molar ratio, extra water content, and sodium hydroxide concentration were identified as the most critical parameters affecting the compressive strength of MK-based geopolymer concrete. This research contributes to advancing the development of sustainable construction materials, streamlining experimental tasks, minimizing the need for labor and materials, improving time efficiency, and providing valuable insights for optimizing the design of MK-based geopolymer concrete.

Analysis of factors influencing the energy efficiency in Chinese wastewater treatment plants through machine learning and SHapley Additive exPlanations

Wastewater treatment plants (WWTPs) contribute significantly to the control of pollution in water. However, they are significant energy consumers. Identifying the factors influencing energy consumption is crucial for enhancing the energy efficiency of WWTPs. To address this, the unit energy consumption (UEC) of WWTPs was predicted using machine learning models. In order to accurately evaluate WWTPs' energy utilization efficiency, a comprehensive energy evaluation indicator, UEC (kWh/kg TODremoved) was utilized in this study. Among the prediction models, the eXtreme Gradient Boosting (XGBoost) achieves the highest prediction accuracy. SHapley Additive exPlanations (SHAP) was adopted as the model explanation system, and the results revealed that UEC was negatively affected by TN concentration, which was the most influential factor. The stoichiometry-based model calculation result indicates that the nitrification consumes average 77 % of the overall oxygen demand. SHAP analysis illustrated that the UEC of main technologies decreases with increasing influential factors. Partial dependence plot (PDP) compared average UEC of these technologies and SBR consumed the least amount of energy. The research also indicated that low influent TN concentration is the main problem in China. Consequently, it is imperative to exert efforts in ensuring the influent TN concentration while simultaneously making appropriate adjustments to the treatment process. This study provides valuable implications and methods for retrofitting and upgrading WWTPs.

Development and internal validation of machine learning models for personalized survival predictions in spinal cord glioma patients

BACKGROUND CONTEXTNumerous factors have been associated with the survival outcomes in patients with spinal cord gliomas (SCG). Recognizing these specific determinants is crucial, yet it is also vital to establish a reliable and precise prognostic model for estimating individual survival outcomes. OBJECTIVEThe objectives of this study are twofold: first, to create an array of interpretable machine learning (ML) models developed for predicting survival outcomes among SCG patients; and second, to integrate these models into an easily navigable online calculator to showcase their prospective clinical applicability. STUDY DESIGNThis was a retrospective, population-based cohort study aiming to predict the outcomes of interest, which were binary categorical variables, in SCG patients with ML models. PATIENT SAMPLEThe National Cancer Database (NCDB) was utilized to identify adults aged 18 years or older who were diagnosed with histologically confirmed SCGs between 2010 and 2019. OUTCOME MEASURESThe outcomes of interest were survival outcomes at three specific time points postdiagnosis: 1, 3, and 5 years. These outcomes were formed by combining the “Vital Status” and “Last Contact or Death (Months from Diagnosis)” variables. Model performance was evaluated visually and numerically. The visual evaluation utilized receiver operating characteristic (ROC) curves, precision-recall curves (PRCs), and calibration curves. The numerical evaluation involved metrics such as sensitivity, specificity, accuracy, area under the PRC (AUPRC), area under the ROC curve (AUROC), and Brier Score. METHODSWe employed five ML algorithms—TabPFN, CatBoost, XGBoost, LightGBM, and Random Forest—along with the Optuna library for hyperparameter optimization. The models that yielded the highest AUROC values were chosen for integration into the online calculator. To enhance the explicability of our models, we utilized SHapley Additive exPlanations (SHAP) for assessing the relative significance of predictor variables and incorporated partial dependence plots (PDPs) to delineate the influence of singular variables on the predictions made by the top performing models. RESULTSFor the 1-year survival analysis, 4,913 patients [5.6% with 1-year mortality]; for the 3-year survival analysis, 4,027 patients (11.5% with 3-year mortality]; and for the 5-year survival analysis, 2,854 patients (20.4% with 5-year mortality) were included. The top models achieved AUROCs of 0.938 for 1-year mortality (TabPFN), 0.907 for 3-year mortality (LightGBM), and 0.902 for 5-year mortality (Random Forest). Global SHAP analyses across survival outcomes at different time points identified histology, tumor grade, age, surgery, radiotherapy, and tumor size as the most significant predictor variables for the top-performing models. CONCLUSIONSThis study demonstrates ML techniques can develop highly accurate prognostic models for SCG patients with excellent discriminatory ability. The interactive online calculator provides a tool for assessment by physicians (https://huggingface.co/spaces/MSHS-Neurosurgery-Research/NCDB-SCG). Local interpretability informs prediction influences for a given individual. External validation across diverse datasets could further substantiate potential utility and generalizability. This robust, interpretable methodology aligns with the goals of precision medicine, establishing a foundation for continued research leveraging ML's predictive power to enhance patient counseling.

Rehabilitation in the intensive care unit: How amount of physical and occupational therapy affects patients' function and hospital length of stay.

Patients in the intensive care unit (ICU) often experience extended periods of immobility. Following hospital discharge, many face impaired mobility and never return to their baseline function. Although the benefits of physical and occupational rehabilitation are well established in non-ICU patients, a paucity of work describes effective practices to alleviate ICU-related declines in mobility. To assess how rehabilitation with physical and occupational therapy (PT-OT) during ICU stays affects patients' mobility, self-care, and length of hospital stay. Retrospective cohort study. Inpatient ICU. A total of 6628 adult patients who received physical rehabilitation across multiple sites (Arizona, Florida, Minnesota, and Wisconsin) of a single institution between January 2018 and December 2021. Not applicable. Descriptive statistics, linear regression models, and gradient boosting machine methods were used to determine the relationship between the amount of PT-OT received and outcomes of hospital length of stay (LOS), Activity Measure for Post-Acute Care Daily Activity and Basic Mobility scores. The 6628 patients who met inclusion criteria received an average (median) of 23 (range: 1-89) minutes of PT-OT per day. Regression analyses showed each additional 10 minutes of PT-OT per day was associated with a 1.0% (95% confidence interval [CI]: 0.41-1.66, p < .001) higher final Basic Mobility score, a 1.8% (95% CI: 1.30%-2.34%, p < .001) higher final Daily Activity score, and a 1.2-day (95% CI: -1.28 to -1.09, p < .001) lower hospital LOS. One-dimensional partial dependence plots revealed an exponential decrease in predicted LOS as minutes of PT-OT received increased. Higher rehabilitation minutes provided to patients in the ICU may reduce the LOS and improve patients' functional outcomes at discharge. The benefits of rehabilitation increased with increasing amounts of time of therapy received.