Unraveling nonlinear impacts of seasonal climate and built environments on exercise walking in high-density cities via a modified machine learning approach

BackgroundTotal joint arthroplasty (TJA) complications necessitate the development of accurate risk prediction models; however, interpretability in machine learning remains a challenge. While Shapley Additive Explanations (SHAP) offers insights at the individual level, partial dependence plots (PDPs) may provide a better understanding at the population level for developing clinical guidelines. This study compared PDPs and SHAP in explaining machine learning-based 30-day complication risk prediction following TJA.MethodsWe conducted a retrospective cohort study using the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) database (2019–2023), including 517,826 primary TJA cases. Binary classification models (Random Forest, Gradient Boosting) predicted composite 30-day complications based on 20 clinical predictors. A comprehensive interpretability analysis employed directional concordance validation between PDP and SHAP, permutation importance thresholding (5% relative influence), followed by one- and two-dimensional partial dependence analyses with explicit interaction modeling.ResultsThe cohort comprised 517,826 primary TJA procedures with a complication rate of 6.67%. The baseline Random Forest model achieved test AUC = 0.678. Directional concordance analysis demonstrated 97.8% weighted agreement between PDP trends and SHAP attributions, validating methodological comparison. Threshold analysis identified seven significant features, with interaction effects accounting for 49.9% of total model influence (71.9% among top features). PDPs showed actionable dose–response relationships, including critical thresholds for preoperative hematocrit (< 38%), operative time (> 120 min), and complementary interactions, such as age × ASA classification (19.1% importance), operative time × ASA classification (10.1%), and hematocrit × diabetes (6.4%). Comparative patient analysis demonstrated that while SHAP quantified individual contributions, only PDPs provided population thresholds directly translatable to institutional protocols.ConclusionPDPs appear more methodologically appropriate than SHAP for population-level clinical guideline development, offering actionable dose–response relationships and population risk thresholds that SHAP’s individualized attribution framework cannot provide. The dominance of interaction effects among the most influential predictors validates that PDPs accurately capture complementary relationships while presenting them in a format directly applicable to evidence-based perioperative protocols and institutional quality improvement initiatives.Video Supplementary InformationThe online version contains supplementary material available at 10.1186/s42836-025-00360-9.

Climate signals, driven by complex interactions and nonlinear relationships, shape weather patterns and long-term trends, complicating the identification of dominant drivers due to collinearity. This study investigates the consistency and uncertainty of machine learning (ML) techniques for feature importance in climate science, comparing SHapley Additive exPlanations (SHAP), Partial Dependence Plots (PDPs), and gain-based feature importance from Extreme Gradient Boosting (XGBoost). SHAP’s integration with Feed Forward Neural Networks (FFNN) and XGBoost is evaluated to assess model-specific uncertainties. Using winter precipitation data from Ohio, USA, as a case study, the relative contributions of global warming (GW) and the Interdecadal Pacific Oscillation (IPO) to precipitation changes are quantified. Results show GW consistently ranks higher than IPO in at least 60% of stations across all methods, with SHAP and PDPs agreeing in 89% of stations. Global SHAP importance from FFNN and XGBoost aligns in 82% of stations, with GW contributing 15% more than IPO on average, though disagreements in 18% of stations highlight model-dependent uncertainties. Temporal analysis using SHAP values indicates a moderate discrepancy in feature importance between FFNN and XGBoost models (Pearson correlation ≈ 0.5), despite their consensus on the increasing dominance of GW in recent decades, contributing to wetter winters. Regression analysis further confirms that GW accounts for approximately 70% of the multi-decadal variability in winter precipitation across Ohio, with PDPs indicating a strong monotonicity (ρ = 0.94) between warming levels and precipitation increase. PDPs visualize marginal effects but struggle with interactions, while gain-based methods tend to favor features with a greater number of effective split points that reduce loss. SHAP, though robust for ranking, varies with the base model. An ensemble framework is proposed, demonstrating the value of combining these ML techniques complementarily to account for uncertainties and enhance interpretability. This study highlights the importance of addressing methodological uncertainties in feature importance rankings to provide robust insights for climate modeling.

In recent years, with the constant change in the global climate, the effect of climate factors on net primary productivity (NPP) has become a hot research topic. However, two opposing views have been presented in this research area: global NPP increases with global warming, and global NPP decreases with global warming. The main reasons for these two opposite results are the tremendous differences among seasonal and annual climate variables, and the growth of plants in accordance with these climate variables. Therefore, it will fail to fully clarify the relation between vegetation growth and climate changes by research that relies solely on annual data. With seasonal climate variables, we may clarify the relation between vegetation growth and climate changes more accurately. Our research examined the arid and semiarid areas in China (ASAC), which account for one quarter of the total area of China. The ecological environment of these areas is fragile and easily affected by human activities. We analyzed the influence of climate changes, especially the changes in seasonal climate variables, on NPP, with Climatic Research Unit (CRU) climatic data and Moderate Resolution Imaging Spectroradiometer (MODIS) satellite remote data, for the years 2000–2010. The results indicate that: for annual climatic data, the percentage of the ASAC in which NPP is positively correlated with temperature is 66.11%, and 91.47% of the ASAC demonstrates a positive correlation between NPP and precipitation. Precipitation is more positively correlated with NPP than temperature in the ASAC. For seasonal climatic data, the correlation between NPP and spring temperature shows significant regional differences. Positive correlation areas are concentrated in the eastern portion of the ASAC, while the western section of the ASAC generally shows a negative correlation. However, in summer, most areas in the ASAC show a negative correlation between NPP and temperature. In autumn, precipitation is less important in the west, as opposed to the east, in which it is critically important. Temperatures in winter are a limiting factor for NPP throughout the region. The findings of this research not only underline the importance of seasonal climate variables for vegetation growth, but also suggest that the effects of seasonal climate variables on NPP should be explored further in related research in the future.

PurposeThe purpose of this study is to analyze the seasonal spatiotemporal climate variability in the Borena zone of Ethiopia and its effects on agriculture and livestock production. By examining these climate variables in relation to global sea surface temperatures (SST) and atmospheric pressure systems, the study seeks to understand the underlying mechanisms driving local climate variability. Furthermore, it assesses how these climate variations impact crop yields, particularly wheat and livestock production, providing valuable insights for developing effective adaptation strategies and policies to enhance food security and economic stability in the region.Design/methodology/approachThe design and methodology of this study involve a multifaceted approach to analyzing seasonal spatiotemporal climate variability in the Borena zone of Ethiopia. The research uses advanced statistical techniques, including rotated empirical orthogonal function (EOF) and rotated principal component analysis (RPCA), to identify and quantify significant patterns in seasonal rainfall, temperature and drought indices over the period from 1981 to 2022. These methods are used to reveal the spatiotemporal variations and trends in climate variables. To understand the causal mechanisms behind these variations, the study correlates seasonal rainfall data with global SST and examines atmospheric pressure systems and wind vectors. In addition, the impact of climate variability on agricultural and livestock production is assessed by linking observed climate patterns with changes in crop yields, particularly wheat and livestock productivity. This comprehensive approach integrates statistical analysis with environmental and agricultural data to provide a detailed understanding of climate dynamics and their practical implications.FindingsThe findings of this study reveal significant seasonal spatiotemporal climate variability in the Borena zone of Ethiopia, characterized by notable patterns and trends in rainfall, temperature and drought indices from 1981 to 2022. The analysis identified that over 84% of the annual rainfall occurs during the March to May (MAM) and September to November (SON) seasons, with MAM contributing approximately 53% and SON over 31%, highlighting these as the primary rainfall periods. Significant spatiotemporal variations were observed, with northwestern (35.4%), southern (34.9%) and northeastern (19.3%) are dominant variability parts of the zone during MAM season, similarly southeastern (48.7%), and northcentral (37.8%) are dominant variability parts of the zone during SON season. Trends indicating that certain subregions experience more pronounced changes in climate variables in both seasons. Correlation with global SST and an examination of atmospheric pressure systems elucidated the mechanisms driving these variations, with significant correlation with the southern and central part of Indian Ocean. This study also found that fluctuations in climate variables significantly impact crop production, particularly wheat and livestock productivity in the region, underscoring the need for adaptive strategies to mitigate adverse effects on agriculture and food security.Research limitations/implicationsThe implications of this study highlight the need for robust adaptation strategies to mitigate the effects of climate variability. Detailed research on seasonal climate patterns and the specific behaviors of livestock and crops is essential. Gaining a thorough understanding of these dynamics is critical for developing resilient adaptation strategies tailored to the unique ecological and economic context of the Borana zone. Future research should focus on seasonal climate variations and their implications to guide sustainable development and livelihood adjustments in the region.Originality/valueThis study offers significant originality and value by providing a detailed analysis of seasonal spatiotemporal climate variability in the Borena zone of Ethiopia, using advanced statistical techniques such as rotated EOF and RPCA. By integrating these methods with global SST data and atmospheric pressure systems, the research delivers a nuanced understanding of how global climatic factors influence local weather patterns. The study’s novel approach not only identifies key trends and patterns in climate variables over an extensive historical period but also links these findings to practical outcomes in crop and livestock production. This connection is crucial for developing targeted adaptation strategies and policies, thereby offering actionable insights for enhancing agricultural practices and food security in the region. The originality of this work lies in its comprehensive analysis and practical relevance, making it a valuable contribution to both climate science and regional agricultural planning.

Optimizing subway-based transit-oriented development (TOD) requires strategic alignment of transit infrastructure, land use balance, and multimodal accessibility. Shapley additive explanations (SHAP) and partial dependence plot (PDP) analyses are used to identify critical saturation thresholds for station-level TOD parameters. The results reveal significant variations in TOD effectiveness on the basis of subway network density in the urban context of five major cities in South Korea. In high-density cities, subway efficiency is primarily related to multimodal integration, particularly efficient bus-subway transfers and pedestrian connectivity. In medium-density cities, a poor land use mix negatively impacts ridership, highlighting the need for balanced station-area development. In hybrid-density cities, subway efficiency depends on optimal station exit configurations and controlled parking availability rather than multimodal connectivity alone. SHAP analysis is used to identify the key determinants of subway ridership, whereas PDP analysis is applied to determine saturation effects, revealing optimal thresholds related to station exits, parking availability, and land use mix. These thresholds can help planners avoid inefficient overinvestment. The findings emphasize the necessity of city-specific TOD strategies, indicating that uniform national policies yield suboptimal outcomes. This study provides data-driven insights that enable planners to align subway investments with regional mobility demands, ensuring sustainable urban transit development.

BackgroundMachine learning (ML) model is increasingly used to predict short-term outcome in critically ill patients, but the study for long-term outcome is sparse. We used explainable ML approach to establish 30-day, 90-day and 1-year mortality prediction model in critically ill ventilated patients.MethodsWe retrospectively included patients who were admitted to intensive care units during 2015–2018 at a tertiary hospital in central Taiwan and linked with the Taiwanese nationwide death registration data. Three ML models, including extreme gradient boosting (XGBoost), random forest (RF) and logistic regression (LR), were used to establish mortality prediction model. Furthermore, we used feature importance, Shapley Additive exPlanations (SHAP) plot, partial dependence plot (PDP), and local interpretable model-agnostic explanations (LIME) to explain the established model.ResultsWe enrolled 6994 patients and found the accuracy was similar among the three ML models, and the area under the curve value of using XGBoost to predict 30-day, 90-day and 1-year mortality were 0.858, 0.839 and 0.816, respectively. The calibration curve and decision curve analysis further demonstrated accuracy and applicability of models. SHAP summary plot and PDP plot illustrated the discriminative point of APACHE (acute physiology and chronic health exam) II score, haemoglobin and albumin to predict 1-year mortality. The application of LIME and SHAP force plots quantified the probability of 1-year mortality and algorithm of key features at individual patient level.ConclusionsWe used an explainable ML approach, mainly XGBoost, SHAP and LIME plots to establish an explainable 1-year mortality prediction ML model in critically ill ventilated patients.

IntroductionThe need for eXplainable Artificial Intelligence (XAI) in healthcare is more critical than ever, especially as regulatory frameworks such as the European Union Artificial Intelligence (EU AI) Act mandate transparency in clinical decision support systems. Post hoc XAI techniques such as Local Interpretable Model-Agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDPs) are widely used to interpret Machine Learning (ML) models for disease risk prediction, particularly in tabular Electronic Health Record (EHR) data. However, their reliability under real-world scenarios is not fully understood. Class imbalance is a common challenge in many real-world datasets, but it is rarely accounted for when evaluating the reliability and consistency of XAI techniques.MethodsIn this study, we design a comparative evaluation framework to assess the impact of class imbalance on the consistency of model explanations generated by LIME, SHAP, and PDPs. Using UK primary care data from the Clinical Practice Research Datalink (CPRD), we train three ML models: XGBoost (XGB), Random Forest (RF), and Multi-layer Perceptron (MLP), to predict lung cancer risk and evaluate how interpretability is affected under class imbalance when compared against a balanced dataset. To our knowledge, this is the first study to evaluate explanation consistency under class imbalance across multiple models and interpretation methods using real-world clinical data.ResultsOur main finding is that class imbalance in the training data can significantly affect the reliability and consistency of LIME and SHAP explanations when evaluated against models trained on balanced data. To explain these empirical findings, we also present a theoretical analysis of LIME and SHAP to understand why explanations change under different class distributions. It is also found that PDPs exhibit noticeable variation between models trained on imbalanced and balanced datasets with respect to clinically relevant features for predicting lung cancer risk.DiscussionThese findings highlight a critical vulnerability in current XAI techniques, i.e., their interpretability are significantly affected under skewed class distributions, which is common in medical data and emphasises the importance of consistent model explanations for trustworthy ML deployment in healthcare.

Validity of Walk Score® as a measure of neighborhood walkability in Japan

Data visualization performs a vital function in enhancing the interpretability of machine learning fashions, addressing the "black box" nature of complex algorithms. As the system getting to know models come to be more and more state-of-the-art, know-how their selection-making methods will become greater challenging. Visualizations offer an intuitive manner to get to the bottom of the complex relationships within those models, providing insights into feature significance, model conduct, and ability biases. Techniques including partial dependence plots, LIME (Local Interpretable Model-agnostic Explanations), and SHAP (SHapley Additive exPlanations) values serve as powerful gear to visualize and interpret system getting to know predictions. These visualizations not best aid statistics scientists in debugging and refining models however additionally make contributions to building agree with and transparency, essential factors for broader attractiveness and deployment of machine learning answers in real-global programs. As the field of system learning progresses, exploring novel and effective methods to visualize model interpretability turns into vital for empowering each expert and non-professionals to recognise, consider, and correctly use the insights derived from complex device studying structures. Data visualization is a crucial element in unraveling the complex layers of devices, gaining knowledge of fashions, and improving their interpretability. As these models grow in complexity, expertise in their choice-making mechanisms turns into paramount for agree with, responsibility, and effective deployment. Visual representations function a powerful device to distill complex records into handy codecs, permitting stakeholders to recognise and scrutinize the version's conduct.Techniques which include function importance plots, partial dependence plots, and SHAP (SHapley Additive exPlanations) values provide insights into the impact of person features on model predictions. Feature significance plots highlight the significance of each input variable, assisting in the identity of influential elements. Partial dependence plots showcase the relationship between a specific characteristic and the version's output at the same time as keeping different variables steady, offering a nuanced know-how of their impact.SHAP values, alternatively, provide a greater holistic view by means of assigning a contribution score to every feature for every prediction, revealing the collective have an effect on of capabilities at the model's choice. These visualizations allow stakeholders to grasp not most effective which functions are critical but additionally how they have interaction, fostering a more nuanced understanding of the model's choice common sense.Moreover, confusion matrices, ROC curves, and precision-remember curves are quintessential tools for comparing version performance. These visualizations facilitate a comprehensive evaluation of class fashions by illustrating real positives, actual negatives, false positives, and false negatives. ROC curves graphically constitute the change-off among sensitivity and specificity, assisting within the selection of suitable.

Achieving sustainable cities and promoting responsible consumption require innovative approaches to chemical design and manufacturing. Precise prediction of chemical biodegradability is crucial for evaluating environmental concerns and facilitating the transition towards green chemistry. This study investigates the effectiveness of ten distinct groups of three-dimensional (3D) molecular descriptors for classifying compounds with rapid biodegradability. The Merck molecular force field (MMFF94s) was used to compute descriptors and generate 3D conformations for a dataset of chemical compounds. The dataset underwent rigorous preprocessing, including feature selection, outlier management, and scaling. Support Vector Machines (SVMs) were tested alongside three tree-based ensemble learning algorithms: Extreme Gradient Boosting (XGBoost), Gradient Boosting Machine (GBM), and Random Forest. Bayesian optimization was employed to optimize model hyperparameters and enhance cross-validated Area Under the Receiver Operating Characteristic Curve (AUC). The GETAWAY descriptors, 3D autocorrelation descriptors, and 3D-MoRSE descriptors consistently demonstrated superior performance compared to other descriptors across all machine learning models. An SVM model trained on 3D autocorrelation descriptors achieved the highest prediction accuracy (0.88), sensitivity (0.83), specificity (0.91), F1-score (0.82), Cohen’s Kappa statistic (0.74), and an AUC of 0.93 on an independent test set. Advanced analytical techniques, including Permutation Feature Importance (PFI), SHapley Additive exPlanations(SHAP), and partial dependency plots (PDP) were utilized to identify the most influential 3D autocorrelation descriptors. The findings of this study demonstrate that 3D molecular descriptors, particularly 3D autocorrelations, play a critical role in developing accurate and interpretable models for predicting chemical biodegradability. These models contribute significantly to the advancement of green chemical design and the development of effective regulatory policies that support the objectives of SDG 11 (Sustainable Cities and Communities) and SDG 12 (Responsible Consumption and Production). By fostering sustainable chemical manufacturing practices, we can create healthier and more resilient urban environments while minimizing the environmental impact of human activities.

This study investigates the effectiveness of inclined double cutoff walls installed beneath hydraulic structures by employing five machine learning models: Random Forest(RF), Adaptive Boosting(AdaBoost), eXtreme Gradient Boosting(XGBoost), Light Gradient Boosting Machine(LightGBM), and Categorical Boosting (CatBoost). A comprehensive dataset of 630 samples was gathered from previous studies, including key input variables such as the relative distance between the cutoff wall and the structure's apron width (L/B), the inclination angle ratio between downstream and upstream cutoffs (θ2/θ1), the depth ratio of downstream to upstream cutoff walls (d2/d1), and the relative downstream cutoff depth to the permeable layer depth (d2/D). Outputs considered were the relative uplift force (U/Uo), the relative exit hydraulic gradient (iR/iRo), and the relative seepage discharge per unit structure length (q/qo). The dataset was split with a 70:30 ratio for training and testing. Hyperparameter optimization was conducted using Bayesian Optimization (BO) coupled with five-fold cross-validation to enhance model performance. Results showed that the CatBoost model demonstrated superior performance over other models, consistently yielding high R2 values, specifically surpassing 0.95, 0.93, and 0.97 for U/Uo, iR/iRo, and q/qo, respectively, along with low RMSE scores below 0.022, 0.089, and 0.019 for the same variables. A feature importance analysis is conducted using SHapley Additive exPlanations(SHAP) and Partial Dependence Plot (PDP). The analysis revealed that L/B was the most influential predictor for U/Uo and iR/iRo, while d2/D played a crucial role in determining q/qo. Moreover, PDPs illustrated a positive linear relationship between L/B and U/Uo, a V-shaped impact of d2/d1 on iR/iRo and q/qo, and complex nonlinear interactions for θ2/θ1 across all target variables. Furthermore, an interactive Graphical User Interface(GUI) was developed, enabling engineers to efficiently predict output variables and apply model insights in practical scenarios.

Controlling seawater intrusion (SWI) into freshwater aquifers is crucial for preserving water quality in coastal groundwater management. This research evaluates the performance of three machine learning (ML) models: eXtreme Gradient Boosting (BO-XGB), Light Gradient Boosting Machine (BO-LGB), and Categorical Gradient Boosting (BO-CGB) in predicting the SWI wedge length. A database of 345 numerical simulations was compiled from previous research, and Bayesian Optimization (BO) with fivefold cross-validation was used to fine-tune the models. The inputs included abstraction well distance (Xa), abstraction well depth (Ya), recharge well distance (Xr), recharge well depth (Yr), abstraction rate (Qa), artificial recharge rate (Qr), and SWI wedge length (L). Results show that BO-CGB consistently achieved the best performance, with high R2 values (0.996 in training and 0.969 in testing) and low RMSE values (0.439 m in training and 1.327 m in testing). SHapley Additive exPlanations (SHAP) analysis highlighted that Qa and Qr had the most significant impact on SWI wedge length predictions, followed by Xa and Ya. Partial Dependence Plot (PDP) analysis revealed a strong negative correlation between flow variables Qa and Qr and wedge length, while Xr displayed a more complex, non-linear pattern. BO-CGB emerged as the most reliable model for predicting SWI wedge length. To facilitate practical application, an interactive Graphical User Interface (GUI) was developed, enabling users to input variables and receive instant predictions, enhancing the practical usability of the ML models in managing SWI in coastal aquifers.

Software Effort estimation (SEE) is a vital task for project management as it is essential for resource allocation and project planning. Numerous algorithms have been investigated for forecasting software effort, yet achieving precise predictions remains a significant hurdle in the software industry. To achieve optimal accuracy, machine learning algorithms are employed. Remarkably, Random Forest (RF) algorithm produced better accuracy when compared with various algorithms. In this paper, the prediction is extended by increasing the number of trees and Improved Random Forest (IRF) is implemented by including three decision techniques such as residual analysis, partial dependence plots and feature engineering to improve prediction accuracy. To make improved random forest to be adaptive, it is further extended in this paper by integrating three techniques such as: Bayesian Optimization with Deep Kernel Learning (BO-DKL) to adaptively set hyperparameters, Time-Series Residual Analysis to detect autocorrelation patterns among model error, and Explainable AI techniques Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) to improve feature interpretability. This Improved Adaptive Random Forest (IARF) mutually contributes to a comprehensive evaluation and improvement of accuracy in prediction. Metrics used for evaluation are Mean Absolute Error (MAE), Root Mean Square Error (RMSE), R-Squared, Mean Absolute Percentage Error (MAPE), Mean Absolute Scaled Error (MASE) and Prediction Interval Coverage Probability (PICP). Overall, the improved adaptive RF model had an average improvement ratio of 18.5% on MAE, 20.3% on RMSE, 3.8% on R2, 5.4% on MAPE, 7% reduction in MASE and a 3-5% improvement in PICP across all data sets compared to the Random Forest model, with much improved prediction accuracy. These findings validate that the combination of adaptive learning methods and explainability-based adjustments considerably improves accuracy of software effort estimation models and facilitates more trustworthy decision-making in software development projects.

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

BackgroundSince the inconspicuous nature of early signs associated with Chronic Obstructive Pulmonary Disease (COPD), individuals often remain unidentified, leading to suboptimal opportunities for timely prevention and treatment. The purpose of this study was to create an explainable artificial intelligence framework combining data preprocessing methods, machine learning methods, and model interpretability methods to identify people at high risk of COPD in the smoking population and to provide a reasonable interpretation of model predictions.MethodsThe data comprised questionnaire information, physical examination data and results of pulmonary function tests before and after bronchodilatation. First, the factorial analysis for mixed data (FAMD), Boruta and NRSBoundary-SMOTE resampling methods were used to solve the missing data, high dimensionality and category imbalance problems. Then, seven classification models (CatBoost, NGBoost, XGBoost, LightGBM, random forest, SVM and logistic regression) were applied to model the risk level, and the best machine learning (ML) model’s decisions were explained using the Shapley additive explanations (SHAP) method and partial dependence plot (PDP).ResultsIn the smoking population, age and 14 other variables were significant factors for predicting COPD. The CatBoost, random forest, and logistic regression models performed reasonably well in unbalanced datasets. CatBoost with NRSBoundary-SMOTE had the best classification performance in balanced datasets when composite indicators (the AUC, F1-score, and G-mean) were used as model comparison criteria. Age, COPD Assessment Test (CAT) score, gross annual income, body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), anhelation, respiratory disease, central obesity, use of polluting fuel for household heating, region, use of polluting fuel for household cooking, and wheezing were important factors for predicting COPD in the smoking population.ConclusionThis study combined feature screening methods, unbalanced data processing methods, and advanced machine learning methods to enable early identification of COPD risk groups in the smoking population. COPD risk factors in the smoking population were identified using SHAP and PDP, with the goal of providing theoretical support for targeted screening strategies and smoking population self-management strategies.