Interpretable Machine Learning Models Research Articles

Enhancement of atmospheric oxidation capacity induced co-pollution of the O3 and PM2.5 in Lanzhou, northwest China

In recent years, the co-pollution of surface ozone (O3) and fine particulate matter (PM2.5) has emerged as a critical concern within specific regions of China's atmospheric environment. This study employed a comprehensive approach by integrating statistical analysis with the interpretable ensemble machine learning model. Delving deeply into the intricate mechanisms behind O3 and PM2.5 co-pollution in Lanzhou city from 2019 to 2022, the research synthesized and analyzed an array of data sources, including ground observations, a multi-parameter lidar system, and meteorological data. Our findings, derived from ground observations to vertical distribution, unequivocally confirm that the enhancement of atmospheric oxidation capacity serves as a critical driver in the genesis of secondary particles, playing a substantial role in the augmented levels of O3 and PM2.5 experienced during the warm season. Moreover, the impact of local weather patterns is indispensable as it precipitates a relatively stable mid-level atmosphere, culminating in elevated surface concentrations of both PM2.5 and O3. Overall, this study emphatically underscores the importance of adopting a comprehensive approach to address these environmental challenges.

Forecasting carbon price trends based on an interpretable light gradient boosting machine and Bayesian optimization

The future carbon price is crucial to relevant companies, investors, and carbon policymakers, and the significance of carbon price prediction research is self-evident. However, existing study usually predicts actual carbon prices, rarely considering price trends and lacking reasonable interpretations for the prediction model. Thus, in this study, an interpretable machine learning model is proposed to predict carbon price trends. It integrates five methods, including the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), two-stage feature selection (TFS), light gradient boosting machine (LightGBM) optimized by Bayesian optimization algorithm (BOA), and SHapley Additive exPlanations (SHAP). The effectiveness of the proposed model is validated with the carbon prices of the Hubei carbon trading market, which has the largest volume among Chinese markets. The experimental results showed that the proposed model outperforms other benchmark models under five evaluation criteria, including AUC, Accuracy, Precision, Recall, and F1 score, on multiple-step predictions. For one-step-ahead prediction, the average hit ratio results are 0.8342, 77.32 %, 77.87 %, 76.83 %, and 76.88 % respectively; for five-step-ahead prediction, the average hit ratio results are 0.7641, 69.25 %, 71.17 %, 71.97 %, and 71.00 % respectively; and for ten-step-ahead prediction, the average hit ratio results are 0.7519, 69.11 %, 73.80 %, 69.61 %, and 71.16 % respectively. The SHAP model interpretation results indicated that the high-frequency intrinsic mode function (IMF) components of the historical carbon price are the most important features for predicting carbon price trends. This study contributes by forecasting both the upward and downward trends of carbon prices through multi-step-ahead forecasting with the LightGBM model and further interpreting the model’s predictions with the SHAP approach. Therefore, the proposed model has excellent forecasting performance with interpretability, which is an effective tool for forecasting carbon price trends.

Interpretable machine learning model using Extreme Gradient Boosting with Nonnegative Matrix Factorization improves the accuracy of arrhythmic risk prediction in Brugada Syndrome

Abstract Background Both depolarization and repolarization abnormalities contribute to ventricular arrhythmogenesis in Brugada syndrome (BrS). In this study, we tested the hypothesis that incorporating latent features extracted by various nonnegative matrix factorization (NML) techniques into an interpretable machine learning (IML) prediction model can outperform IML models without latent variables and logistic regression model. Methods This study was based on a published anonymised dataset of BrS patients from the Hong Kong, China. XGBoost was selected as the IML model, with and without incorporating latent features using 11 different NMF techniques: Bayesian nonnegative matrix factorization (BNMF), Iterated Conditional Modes nonnegative matrix factorization (ICM), Fisher Nonnegative Matrix Factorization for learning Local features (LFNMF), Alternating Nonnegative Least Squares Matrix Factorization Using Projected Gradient (bound constrained optimization) method for each subproblem (LSNMF), Non-smooth Nonnegative Matrix Factorization (NSNMF), Probabilistic Nonnegative Matrix Factorization (PMF), Probabilistic Sparse Matrix Factorization (PSMF), Sparse Nonnegative Matrix Factorization (SNMF) based on alternating nonnegativity constrained least squares, Sparse Network-Regularized Multiple Nonnegative Matrix Factorization (SNMNMF), Penalized Matrix Factorization for Constrained Clustering (PMFCC) and Separable Nonnegative Matrix Factorization (SepNMF). Results A total of 548 patients were included (7.3% females, age at diagnosis: 51.0 [38.0-61.0] years old. Of these, 66 suffered from spontaneous ventricular tachyarrhythmias over 84±55 months follow-up. The baseline model using multivariable logistic regression achieved an area under the curve (AUC) of 0.78 [0.72-0.85], which was improved to 0.88 [0.83-0.93] for the IML model without NMF. The AUC was further increased by incorporating additional latent variables extracted using NSNMF (0.95 [0.92-0.98]) and PMFCC (0.94 [0.88-1.00]). Conclusion Incorporation of latent variables by different NMF techniques into an IML prediction model significantly improved the accuracy of risk prediction in BrS.

Interpretable machine learning models for early prediction of acute kidney injury after cardiac surgery

ObjectivePostoperative acute kidney injury (PO-AKI) is a common complication after cardiac surgery. We aimed to evaluate whether machine learning algorithms could significantly improve the risk prediction of PO-AKI.MethodsThe retrospective cohort study included 2310 adult patients undergoing cardiac surgery in a tertiary teaching hospital, China. Postoperative AKI and severe AKI were identified by the modified KDIGO definition. The sample was randomly divided into a derivation set and a validation set based on a ratio of 4:1. Exploiting conventional logistic regression (LR) and five ML algorithms including decision tree, random forest, gradient boosting classifier (GBC), Gaussian Naive Bayes and multilayer perceptron, we developed and validated the prediction models of PO-AKI. We implemented the interpretation of models using SHapley Additive exPlanation (SHAP) analysis.ResultsPostoperative AKI and severe AKI occurred in 1020 (44.2%) and 286 (12.4%) patients, respectively. Compared with the five ML models, LR model for PO-AKI exhibited the largest AUC (0.812, 95%CI: 0.756, 0.860, all P < 0.05), sensitivity (0.774, 95%CI: 0.719, 0.813), accuracy (0.753, 95%CI: 0.719, 0.781) and Youden index (0.513, 95%CI: 0.451, 0.573). Regarding severe AKI, GBC algorithm showed a significantly higher AUC than the other four ML models (all P < 0.05). Although no significant difference (P = 0.173) was observed in AUCs between GBC (0.86, 95%CI: 0.808, 0.902) and conventional logistic regression (0.803, 95%CI: 0.746, 0.852), GBC achieved greater sensitivity, accuracy and Youden index than conventional LR. Notably, SHAP analyses showed that preoperative serum creatinine, hyperlipidemia, lipid-lowering agents and assisted ventilation time were consistently among the top five important predictors for both postoperative AKI and severe AKI.ConclusionLogistic regression and GBC algorithm demonstrated moderate to good discrimination and superior performance in predicting PO-AKI and severe AKI, respectively. Interpretation of the models identified the key contributors to the predictions, which could potentially inform clinical interventions.

Abstract 15033: Development of an Interpretable Machine Learning Model for Predicting Individual Response to Antihypertensive Treatments

Introduction: Accurate prediction of individual response to anti-hypertensive treatment is crucial in enhancing the safety and efficiency of offering appropriate regimens to attain desired blood pressure (BP) goals. Hypothesis: Machine learning (ML) based models could accurately predict the BP reduction of different antihypertensive regimens. Methods: We utilized data from the LIGHT study, a pragmatic, cluster-randomized trial conducted in 94 primary care institutions in China, involving 13146 visits of 5192 patients who were required to attend the clinic for hypertension management every 3 months during an overall study period of 1 year. Clinical and laboratory variables, baseline BP, and anti-hypertensive therapies at baseline and at follow-up were used to develop models for predicting post-treatment individual BP response. Five ML algorithms were applied as candidate models, and mean squared error (MSE) was used to evaluate the model performance ( panel A ). Results: The mean (standard deviation) age of the study population at baseline was 64.2 (14.0) years, and 2207 (42.5%) were female. 29 features selected by the least absolute shrinkage and selection operator method and purposeful selection were used to build the model, including age, sex, weight, waist circumference, BMI, baseline systolic and diastolic BP, smoking status, visit interval, history of coronary heart disease, diabetes, and dyslipidemia, and baseline and prescript antihypertensive drugs. Among all candidate models, the Catboost model had the best performance with a training MSE of 149.6 and a validation MSE of 135.2 ( panel B ). The most important features include baseline BP, age, weight, and BMI ( panel C and D ). Conclusions: We developed an ML-based model for predicting the response to antihypertensive regimens for assisting clinicians to achieve individualized antihypertensive treatment effectively and safely.

Use of machine learning in understanding transport dynamics of land use and public transportation in a developing city

Rapid urbanization has resulted in problems with pollution, traffic congestion, and public transportation in cities worldwide. Much urbanization research disproportionately focuses on cities in developed countries, where data is easier to obtain, to the detriment of cities in developing countries. To help address this gap, our work proposes a machine learning methodology that explores how land use patterns contribute to predicting an aspect of public transportation in growing cities (aggregate public vehicle speeds), with Cauayan City in the Philippines as the test case. We trained gradient-boosting tree models with road properties, day and time, and land use data as features to predict vehicle speeds on various roads in the city. Among various feature combinations used in training, the best model used road, time, and land use features over a range buffer radii of 50 to 1000 m, yielding an R2 of 0.676 (0.003) and a root mean squared error of 6.61 (0.03) kph. Apart from predicting speeds, Shapley's additive explanation (SHAP) values show that proximity to dense residential and commercial areas decreases predicted vehicle speeds. Our approach uses interpretable machine learning models to obtain insights about urban transportation in developing cities.

Novel Correlation for the Solid–Liquid Mass Transfer Coefficient in Stirred Tanks Developed by Interpreting Machine Learning Models Trained on Literature Data

Novel Correlation for the Solid–Liquid Mass Transfer Coefficient in Stirred Tanks Developed by Interpreting Machine Learning Models Trained on Literature Data

Interpretable Machine Learning Model to Predict Bone Cement Leakage in Percutaneous Vertebral Augmentation for Osteoporotic Vertebral Compression Fracture Based on SHapley Additive exPlanations.

Retrospective study. Our objective is to create comprehensible machine learning (ML) models that can forecast bone cement leakage in percutaneous vertebral augmentation (PVA) for individuals with osteoporotic vertebral compression fracture (OVCF) while also identifying the associated risk factors. We incorporated data from patients (n = 425) which underwent PVA. To predict cement leakage, we devised six models based on a variety of parameters. Evaluate and juxtapose the predictive performances relied on measures of discrimination, calibration, and clinical utility. SHapley Additive exPlanations (SHAP) methodology was used to interpret model and evaluate the risk factors associated with cement leakage. The occurrence rate of cement leakage was established at 50.4%. A binary logistic regression analysis identified cortical disruption (OR 6.880, 95% CI 4.209-11.246), the basivertebral foramen sign (OR 2.142, 95% CI 1.303-3.521), the fracture type (OR 1.683, 95% CI 1.083-2.617), and the volume of bone cement (OR 1.198, 95% CI 1.070-1.341) as independent predictors of cement leakage. The XGBoost model outperformed all others in predicting cement leakage in the testing set, with AUC of .8819, accuracy of .8025, recall score of .7872, F1 score of .8315, and a precision score of .881. Several important factors related to cement leakage were drawn based on the analysis of SHAP values and their clinical significance. The ML based predictive model demonstrated significant accuracy in forecasting bone cement leakage for patients with OVCF undergoing PVA. When combined with SHAP, ML facilitated a personalized prediction and offered a visual interpretation of feature importance.

Single Cell Profiling Unravels Plasma Cell Gene Expression Signatures Predictive of Response to Daratumumab Based Therapy, and Prognosis in Multiple Myeloma

Background: Multiple Myeloma (MM) is the second most common hematologic malignancy and remains incurable. Daratumumab (dara) is a potent anti-CD38 monoclonal antibody used for MM treatment, but responses are heterogeneous, and resistance is inevitable. There is hence an urgent need for biomarkers to predict the response to dara. We hypothesized that gene expression profiles (GEP) of tumor subpopulations could be used to predict the response to dara-based therapy. Methodology: We used the Enhanced Single-Cell Analysis with Protein Expression (ESCAPE) RNA-Seq platform (Singleron Biotechnologies) to simultaneously profile the gene and protein expression of plasma cells from the bone marrow of 15 MM patients prior to treatment with dara-based regimens. These data were combined with publicly available single-cell data from Cohen et al. 2021, resulting in a total of 32 samples. Response was defined based on the international myeloma working group (IMWG) criteria, considering patients achieving a very good partial response (VGPR) or better as “responders,” while those achieving a partial response (PR) or worse were deemed “non-responders”. The data were split into 23 training samples (15 responders and 8 non-responders) and 9 validation samples (3 responders and 6 non-responders), each set containing a combination of cases from the two sources of single-cell data. An interpretable machine-learning model was then built to predict the response of each individual cell. The median response score for each sample was used to predict the patient-level response. The Shapley Additive Explanations (SHAP) values from the dara response prediction model were then used as a starting point for creating a general prognostic model. The 200 most influential genes from the dara response prediction model were used to create a linear response model based on the TCGA (MMRF-Compass) bulk RNAseq dataset. The bulk RNAseq dataset was split 80:20 into training and test datasets. Results from the test dataset are presented here. Results: We analyzed single-cell data from a total of 13,094 plasma cells from 32 BM samples. These included 17 cases of relapsed MM described by Cohen et al., who were treated with dara, carfilzomib, lenalidomide, and dexamethasone. The patients treated at our centre comprised 7 cases of relapsed MM treated with dara, thalidomide, and dexamethasone, and 8 samples from patients with newly diagnosed MM treated with dara, bortezomib, and dexamethasone on clinical trials. All 9 patients in the test dataset had their responses correctly predicted (Fig 1). A separate cohort of non-dara-treated patient samples was also tested using the same model, and the response prediction was not accurate (AUC=0.67), suggesting that this model has specificity for predicting response to dara and is not a general response classifier (data not shown). Based on the SHAP values from the dara response prediction model, we developed a prognostic model from bulk RNAseq data consisting of four genes. A combination of this four-gene signature plus the international staging system (ISS) score resulted in a superior prediction of overall survival compared to the ISS alone with a hazard ratio of 13.5 compared to 2.5 for the ISS alone in the same dataset (Fig 2). Conclusions: Combining single-cell RNA sequencing with machine-learning methods may have value in predicting response to immunotherapy in MM. We propose that gene signatures derived from single-cell data may augment clinical decision-making for determining treatment allocation. We also show that these signatures may be combined with existing prognostic scores, resulting in improved delineation of risk groups. Furthermore, the applicability of our gene signature to a bulk gene expression dataset enhances its clinical relevance. Our study demonstrates the power of single-cell omics to identify novel predictors of response to therapy and prognosis in MM, which may be translated into clinical use.

Interpretable machine learning model based on the systemic inflammation response index and ultrasound features can predict central lymph node metastasis in cN0T1-T2 papillary thyroid carcinoma.

It is arguable whether individuals with T1-T2 papillary thyroid cancer (PTC) who have a clinically negative (cN0) diagnosis should undergo prophylactic central lymph node dissection (pCLND) on a routine basis. Many inflammatory indices, including the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), monocyte-to-lymphocyte ratio (MLR), and systemic immune-inflammatory index (SII), have been reported in PTC. However, the associations between the systemic inflammation response index (SIRI) and the risk of central lymph node metastasis (CLNM) remain unclear. Retrospective research involving 1,394 individuals with cN0T1-T2 PTC was carried out, and the included patients were randomly allocated into training (70%) and testing (30%) subgroups. The preoperative inflammatory indices and ultrasound (US) features were used to train the models. To assess the forecasting factors as well as drawing nomograms, the least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression were utilized. Then eight interpretable models based on machine learning (ML) algorithms were constructed, including decision tree (DT), K-nearest neighbor (KNN), support vector machine (SVM), artificial neural network (ANN), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost). The performance of the models was evaluated by incorporating the area under the precision-recall curve (auPR) and the area under the receiver operating characteristic curve (auROC), as well as other conventional metrics. The interpretability of the optimum model was illustrated via the shapley additive explanations (SHAP) approach. Younger age, larger tumor size, capsular invasion, location (lower and isthmus), unclear margin, microcalcifications, color Doppler flow imaging (CDFI) blood flow, and higher SIRI (≥0.77) were independent positive predictors of CLNM, whereas female sex and Hashimoto thyroiditis were independent negative predictors, and nomograms were subsequently constructed. Taking into account both the auROC and auPR, the RF algorithm showed the best performance, and superiority to XGBoost, CatBoost and ANN. In addition, the role of key variables was visualized in the SHAP plot. An interpretable ML model based on the SIRI and US features can be used to predict CLNM in individuals with cN0T1-T2 PTC.

Investigating the interaction parameters on ventilation supercavitation phenomena: Experimental and numerical analysis with machine learning interpretation

Understanding the optimal values and interactions of parameters within each process is of highest importance. This study is dedicated to exploring the influence of various parameters and their interactions on ventilation supercavitation phenomena through interpretable machine learning (ML) models. In this study, the characteristics of supercavitation on a disk cavitator with enhanced ventilation at different Froude numbers have been examined through both experimental and numerical means. Subsequently, the data generated from the experimental and numerical methods have been employed to create the optimized ML model. Then, to investigate the behavior of important parameters, their interactions with each other, and the resulting impact of these interactions on conditioned cavitation, interpretable machine learning techniques, such as shapley additive explanations, partial dependence plots, and individual conditional expectations, were employed within an optimized ML model. The findings highlight that the ventilation coefficient is the most crucial parameter affecting the characteristics of supercavitation. Ventilation coefficient exhibits a non-linear behavior and performs effectively within the range of 0.06–0.12. Additionally, the water speed parameter and the ratio of the back-body's diameter significantly influence the cavity length and cavitation number. These parameters exhibit complex interactions, shaping the characteristics of blowing supercavitation.

Application of interpretable machine learning models to improve the prediction performance of ionic liquids toxicity

With the wide application prospect of ionic liquids (ILs) as solvent in the future industry, in order to promote green and sustainable chemical engineering, the toxicity problem of common concern has been systematically modeled. Machine learning has promoted the development of chemical property prediction model with its powerful data processing ability. Two typical ensemble learning models, Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), were used to model the toxicity of ILs to Vibrio fischeri in this work. The model's hyperparameters were fine-tuned using Bayesian optimization, and its robustness was enhanced through the 5-fold cross validation. The results of the model comparison showed that the XGBoost model exhibited good generalization ability. In addition, the SHapley Additive exPlanations (SHAP) method was used to explain the model in more detail and the XGBoost model was used to supplement the toxicity value matrix of 1590 ILs.

Creation and interpretation of machine learning models for aqueous solubility prediction

Aim: Solubility prediction is an essential factor in rational drug design and many models have been developed with machine learning (ML) methods to enhance the predictive ability. However, most of the ML models are hard to interpret which limits the insights they can give in the lead optimization process. Here, an approach to construct and interpret solubility models with a combination of physicochemical properties and ML algorithms is presented. Methods: The models were trained, optimized, and tested in a dataset containing 12,983 compounds from two public datasets and further evaluated in two external test sets. More importantly, the SHapley Additive exPlanations (SHAP) and heat map coloring approaches were used to explain the predictive models and assess their suitability to guide compound optimization. Results: Among the different ML methods, random forest (RF) models obtain the best performance in the different test sets. From the interpretability perspective, fragment-based coloring offers a more robust interpretation than atom-based coloring and that normalizing the values further improves it. Conclusions: Overall, for certain applications simple ML algorithms such as RF work well and can outperform more complex methods and that combining them with fragment-coloring can offer guidance for chemists to modify the structure with a desired property. This interpretation strategy is publicly available at https://github.com/Pharmacelera/predictive-model-coloring and could be further applied in other property predictions to improve the interpretability of ML models.

The Price of Explainability in Machine Learning Models for 100-Day Readmission Prediction in Heart Failure: Retrospective, Comparative, Machine Learning Study.

Sensitive and interpretable machine learning (ML) models can provide valuable assistance to clinicians in managing patients with heart failure (HF) at discharge by identifying individual factors associated with a high risk of readmission. In this cohort study, we delve into the factors driving the potential utility of classification models as decision support tools for predicting readmissions in patients with HF. The primary objective of this study is to assess the trade-off between using deep learning (DL) and traditional ML models to identify the risk of 100-day readmissions in patients with HF. Additionally, the study aims to provide explanations for the model predictions by highlighting important features both on a global scale across the patient cohort and on a local level for individual patients. The retrospective data for this study were obtained from the Regional Health Care Information Platform in Region Halland, Sweden. The study cohort consisted of patients diagnosed with HF who were over 40 years old and had been hospitalized at least once between 2017 and 2019. Data analysis encompassed the period from January 1, 2017, to December 31, 2019. Two ML models were developed and validated to predict 100-day readmissions, with a focus on the explainability of the model's decisions. These models were built based on decision trees and recurrent neural architecture. Model explainability was obtained using an ML explainer. The predictive performance of these models was compared against 2 risk assessment tools using multiple performance metrics. The retrospective data set included a total of 15,612 admissions, and within these admissions, readmission occurred in 5597 cases, representing a readmission rate of 35.85%. It is noteworthy that a traditional and explainable model, informed by clinical knowledge, exhibited performance comparable to the DL model and surpassed conventional scoring methods in predicting readmission among patients with HF. The evaluation of predictive model performance was based on commonly used metrics, with an area under the precision-recall curve of 66% for the deep model and 68% for the traditional model on the holdout data set. Importantly, the explanations provided by the traditional model offer actionable insights that have the potential to enhance care planning. This study found that a widely used deep prediction model did not outperform an explainable ML model when predicting readmissions among patients with HF. The results suggest that model transparency does not necessarily compromise performance, which could facilitate the clinical adoption of such models.

XAI Applications in Medical Imaging: A Survey of Methods and Challenges

Medical imaging plays a pivotal role in modern healthcare, aiding in the diagnosis, monitoring, and treatment of various medical conditions. With the advent of Artificial Intelligence (AI), medical imaging has witnessed remarkable advancements, promising more accurate and efficient analysis. However, the black-box nature of many AI models used in medical imaging has raised concerns regarding their interpretability and trustworthiness. In response to these challenges, Explainable AI (XAI) has emerged as a critical field, aiming to provide transparent and interpretable solutions for medical image analysis. This survey paper comprehensively explores the methods and challenges associated with XAI applications in medical imaging. The survey begins with an introduction to the significance of XAI in medical imaging, emphasizing the need for transparent and interpretable AI solutions in healthcare. We delve into the background of medical imaging in healthcare and discuss the increasing role of AI in this domain. The paper then presents a detailed survey of various XAI techniques, ranging from interpretable machine learning models to deep learning approaches with built-in interpretability and post hoc interpretation methods. Furthermore, the survey outlines a wide range of applications where XAI is making a substantial impact, including disease diagnosis and detection, medical image segmentation, radiology reports, surgical planning, and telemedicine. Real-world case studies illustrate successful applications of XAI in medical imaging. The challenges associated with implementing XAI in medical imaging are thoroughly examined, addressing issues related to data quality, ethics, regulation, clinical integration, model robustness, and human-AI interaction. The survey concludes by discussing emerging trends and future directions in the field, highlighting the ongoing efforts to enhance XAI methods for medical imaging and the critical role XAI will play in the future of healthcare. This survey paper serves as a comprehensive resource for researchers, clinicians, and policymakers interested in the integration of Explainable AI into medical imaging, providing insights into the latest methods, successful applications, and the challenges that lie ahead.

Guidelines and Standard Frameworks for AI in Medicine: Protocol for a Systematic Literature Review.

Applications of artificial intelligence (AI) are pervasive in modern biomedical science. In fact, research results suggesting algorithms and AI models for different target diseases and conditions are continuously increasing. While this situation undoubtedly improves the outcome of AI models, health care providers are increasingly unsure which AI model to use due to multiple alternatives for a specific target and the "black box" nature of AI. Moreover, the fact that studies rarely use guidelines in developing and reporting AI models poses additional challenges in trusting and adapting models for practical implementation. This review protocol describes the planned steps and methods for a review of the synthesized evidence regarding the quality of available guidelines and frameworks to facilitate AI applications in medicine. We will commence a systematic literature search using medical subject headings terms for medicine, guidelines, and machine learning (ML). All available guidelines, standard frameworks, best practices, checklists, and recommendations will be included, irrespective of the study design. The search will be conducted on web-based repositories such as PubMed, Web of Science, and the EQUATOR (Enhancing the Quality and Transparency of Health Research) network. After removing duplicate results, a preliminary scan for titles will be done by 2 reviewers. After the first scan, the reviewers will rescan the selected literature for abstract review, and any incongruities about whether to include the article for full-text review or not will be resolved by the third and fourth reviewer based on the predefined criteria. A Google Scholar (Google LLC) search will also be performed to identify gray literature. The quality of identified guidelines will be evaluated using the Appraisal of Guidelines, Research, and Evaluation (AGREE II) tool. A descriptive summary and narrative synthesis will be carried out, and the details of critical appraisal and subgroup synthesis findings will be presented. The results will be reported using the PRISMA (Preferred Reporting Items for Systematic Review and Meta-Analyses) reporting guidelines. Data analysis is currently underway, and we anticipate finalizing the review by November 2023. Guidelines and recommended frameworks for developing, reporting, and implementing AI studies have been developed by different experts to facilitate the reliable assessment of validity and consistent interpretation of ML models for medical applications. We postulate that a guideline supports the assessment of an ML model only if the quality and reliability of the guideline are high. Assessing the quality and aspects of available guidelines, recommendations, checklists, and frameworks-as will be done in the proposed review-will provide comprehensive insights into current gaps and help to formulate future research directions. DERR1-10.2196/47105.

Interpretable machine learning models for predicting venous thromboembolism in the intensive care unit: an analysis based on data from 207 centers

BackgroundVenous thromboembolism (VTE) is a severe complication in critically ill patients, often resulting in death and long-term disability and is one of the major contributors to the global burden of disease. This study aimed to construct an interpretable machine learning (ML) model for predicting VTE in critically ill patients based on clinical features and laboratory indicators.MethodsData for this study were extracted from the eICU Collaborative Research Database (version 2.0). A stepwise logistic regression model was used to select the predictors that were eventually included in the model. The random forest, extreme gradient boosting (XGBoost) and support vector machine algorithms were used to construct the model using fivefold cross-validation. The area under curve (AUC), accuracy, no information rate, balanced accuracy, kappa, sensitivity, specificity, precision, and F1 score were used to assess the model's performance. In addition, the DALEX package was used to improve the interpretability of the final model.ResultsThis study ultimately included 109,044 patients, of which 1647 (1.5%) had VTE during ICU hospitalization. Among the three models, the Random Forest model (AUC: 0.9378; Accuracy: 0.9958; Kappa: 0.8371; Precision: 0.9095; F1 score: 0.8393; Sensitivity: 0.7791; Specificity: 0.9989) performed the best.ConclusionML models can be a reliable tool for predicting VTE in critically ill patients. Among all the models we had constructed, the random forest model was the most effective model that helps the user identify patients at high risk of VTE early so that early intervention can be implemented to reduce the burden of VTE on the patients.

Predicting geogenic groundwater arsenic contamination risk in floodplains using interpretable machine-learning model

Long-term exposure to geogenic arsenic (As)-contaminated groundwater poses a severe threat to public health problems. Generally, elevated As concentrations have been observed with high amounts of ammonium in groundwater of floodplains. An extreme gradient boosting algorithm was conducted to develop a probability model based on hydrogeochemical data, which predicted the occurrence rates of groundwater As on a regional scale. Results showed that concentrations of NH4+, Eh, K, Cl−, SO42−, and NO3− were powerful predictive variables of As exposure. The model revealed the co-enrichment of As with NH4+, suggesting that the mineralization of nitrogen-containing organic matter promoted the reduction of As-bearing iron-oxides. The predicted distribution of high-As groundwater showed high consistency with known spatial distribution of As contamination, and the model also accurately predicted As concentrations in Jiangbei Plain of China and typical As-affected floodplains of Southeast Asia. The model can serve as a low-cost and rapid virtual sensor for detecting As concentrations in private or newly drilled wells, thereby providing critical information for informed management decisions, environmental protection and public health safety.

Understanding market competition between transportation network companies using big data

As in a typical two-sided market, the competition between transportation network companies (TNCs) can lead to market fragmentation and loss of matching efficiency between passengers and drivers, whereas a monopoly market may result in the dominant TNC abusing its market power. Therefore, whether to encourage or discourage competition between TNCs is a debatable question for cities. Prior studies explored this question mostly through mathematical equilibrium models, but few have comprehensively investigated it based on empirical analysis using real-world data. To fill this gap, this study proposes a framework to measure and analyze the competition between TNCs using the most accessible ride-hailing trip data. Specifically, using an interpretable machine learning model, we investigate how TNCs’ pricing and wage strategies influence their market shares and how competition intensity affects passenger cost and driver income. The results based on large-scale trip records from four TNCs in New York City show that the pricing strategy is more influential than the wage strategy on the market shares and competition intensity. Instead of the top TNC, it is the strategies of challenger TNCs (with sizeable but not the biggest market shares) that affect the competition more. Both the passenger cost and driver income can benefit from competition even after considering the potential loss of matching efficiency, while TNCs’ profits shrink with growing competition intensity. These findings suggest that cities should encourage competition between TNCs, yet within a limit. They add empirical evidence to prior studies and provide implications for regulating TNC competition.

A multiscale generative model to understand disorder in domain boundaries.

A continuing challenge in atomic resolution microscopy is to identify significant structural motifs and their assembly rules in synthesized materials with limited observations. Here, we propose and validate a simple and effective hybrid generative model capable of predicting unseen domain boundaries in a potassium sodium niobate thin film from only a small number of observations, without expensive first-principles calculations or atomistic simulations of domain growth. Our results demonstrate that complicated domain boundary structures spanning 1 to 100 nanometers can arise from simple interpretable local rules played out probabilistically. We also found previously unobserved, significant, tileable boundary motifs that may affect the piezoelectric response of the material system, and evidence that our system creates domain boundaries with the highest configurational entropy. More broadly, our work shows that simple yet interpretable machine learning models could pave the way to describe and understand the nature and origin of disorder in complex materials, therefore improving functional materials design.