Tree-based Regression Models Research Articles

Estimating Residual Oil Saturation in Carbonate Rocks: A Combined Approach of Direct Simulation and Data-Driven Analysis

Summary Estimating residual oil saturation (Sor) post-waterflooding is critical for selecting enhanced oil recovery strategies, further field development, and production prediction. We established a data-driven workflow for evaluating Sor in carbonate samples using microcomputed tomography (μ-CT) images. The two-phase lattice Boltzmann method (LBM) facilitated the flooding simulation on 7,192 μ-CT samples. Petrophysical parameters (features) obtained from pore network modeling (PNM) and feature extraction from μ-CT images were utilized to develop tree-based regression models for predicting Sor. Petrophysical features include porosity, absolute permeability, initial water saturation (Swi), pore size distribution (PSD), throat size distributions (TSD), and surface roughness (Ra) distribution. Our method excludes vugs and macro/nanoporosity, which complicates multiscale simulations—a recognized challenge in modeling carbonate rocks. When subdividing the image into numerous subvolumes, certain subvolumes may contain vugs exceeding the dimensions of the subvolume itself. Hence, these vugs were omitted given the entirety of the image constitutes a vug. Conversely, vugs with dimensions smaller than those of the subvolume were not excluded. Despite scale limitations, our subsampling, supported by substantial data volume, ensures our microscale porosity predictions are statistically reliable, setting a foundation for future studies on vugs and nanoporosity’s impact on simulations. The results show that features obtained from dry-sample images can be used for data-driven Sor prediction. We tested three regression models: gradient boosting (GB), random forest (RF), and extreme gradient boosting (XGBoost). Among these, the optimized GB-based model demonstrated the highest predictive capacity for Sor prediction [R2 = 0.87, mean absolute error (MAE) = 1.87%, mean squared error (MSE) = 0.12%]. Increasing the data set size is anticipated to enhance the models’ ability to capture a broader spectrum of rock properties, thereby improving their prediction accuracy. The proposed predictive modeling framework for estimating Sor in heterogeneous carbonate formations aims to supplement conventional coreflooding tests or serve as a tool for rapid Sor evaluation of the reservoir.

Predictive modelling of laser powder bed fusion of Fe-based nanocrystalline alloys based on experimental data using multiple linear regression analysis

This study harnessed bivariate correlational analysis, multiple linear regression analysis and tree-based regression analysis to examine the relationship between laser process parameters and the final material properties (bulk density, saturation magnetization (Ms), and coercivity (Hc)) of Fe-based nano-crystalline alloys fabricated via laser powder bed fusion (LPBF). A dataset comprising of 162 experimental data points served as the foundation for the investigation. Each data point encompassed five independent variables: laser power (P), laser scan speed (v), hatch spacing (h), layer thickness (t), and energy density (E), along with three dependent variables: bulk density, Ms, and Hc. The bivariate correlational analysis unveiled that bulk density exhibited a significant correlation with P, v, h, and E, whereas Ms and Hc displayed significant correlations exclusively with v and P, respectively. This divergence may stem from the strong influence of microstructure on magnetic properties, which can be impacted not only by the laser process parameters explored in this study but also by other factors such as oxygen levels within the build chamber. Furthermore, our statistical analysis revealed that bulk density increased with rising P, h, and E, while decreased with higher v. Regarding the magnetic properties, a high Ms was achievable through low v, while low Hc resulted from high P. It was concluded that P and v were considered as the primary laser process parameters, influencing h and t due to their control over the melt-pool size. The application of multiple linear regression analysis allowed the prediction of the bulk density by using both laser process parameters and energy density. This approach offered a valuable alternative to time-consuming and costly trial-and-error experiments, yielding a low error of less than 1 % between the mean predicted and experimental values. Although a slightly higher error of approximately 6 % was observed for Ms, a clear association was established between Ms and v, with lower v values corresponding to higher Ms values. Additionally, a further comparison was conducted between multiple linear regression and three tree-based regression models to explore the effectiveness of these approaches.

An Optimal, Power Efficient, Internet of Medical Things Framework for Monitoring of Physiological Data Using Regression Models.

The sensors used in the Internet of Medical Things (IoMT) network run on batteries and need to be replaced, replenished or should use energy harvesting for continuous power needs. Additionally, there are mechanisms for better utilization of battery power for network longevity. IoMT networks pose a unique challenge with respect to sensor power replenishment as the sensors could be embedded inside the subject. A possible solution could be to reduce the amount of sensor data transmission and recreate the signal at the receiving end. This article builds upon previous physiological monitoring studies by applying new decision tree-based regression models to calculate the accuracy of reproducing data from two sets of physiological signals transmitted over cellular networks. These regression analyses are then executed over three different iteration varieties to assess the effect that the number of decision trees has on the efficiency of the regression model in question. The results indicate much lower errors as compared to other approaches indicating significant saving on the battery power and improvement in network longevity.

Spatial Image-Based Walkability Evaluation Using Regression Model

Spatial Image-Based Walkability Evaluation Using Regression Model

High resolution prediction and explanation of groundwater depletion across India

Food production in much of the world relies on groundwater resources. In many regions, groundwater levels are declining due to a combination of anthropogenic extraction, localized meteorological and geological characteristics, and climate change. Groundwater in India is characteristic of this global trend, with an agricultural sector that is highly dependent on groundwater and increasingly threatened by extraction far in excess of recharge. The complexity of inputs makes groundwater depletion highly heterogeneous across space and time. However, modeling this heterogeneity has thus far proven difficult. Using two ensemble tree-based regression models, we predict district level seasonal groundwater dynamics to an accuracy of R 2 = 0.4–0.6 and Pearson correlations between 0.6 and 0.8. Further using two high-resolution feature importance methods, we demonstrate that atmospheric humidity, groundwater groundwater-based irrigation, and crop cultivation are the most important predictors of seasonal groundwater dynamics at the district level in India. We further demonstrate a shift in the predictors of groundwater depletion over 1998–2014 that is robustly found between the two feature importance methods, namely increasing importance of deep-well irrigation in Central and Eastern India. These areas coincide with districts where groundwater depletion is most severe. Further analysis shows decreases in crop yields per unit of irrigation over those regions, suggesting decreasing marginal returns for largely increasing quantities of groundwater irrigation used. This analysis demonstrates the public policy value of machine learning models for providing high spatiotemporal accuracy in predicting groundwater depletion, while also highlighting how anthropogenic activity impacts groundwater in India, with consequent implications for productivity and well-being.

Using the surface scattering mechanism from dual-pol SAR data to estimate topsoil particle-sizefractions

Data extracted from Synthetic Aperture Radar (SAR) have been widely employed to estimate soil properties. However, these studies are typically constrained to bare soil conditions, as soil information retrieval in vegetated areas remains challenging. Polarimetric decomposition has emerged as a potentially useful method to separate the scattering contributions of different targets (e.g. canopy/leaves and the underlying soil), which is of significance for areas that are near-permanently covered in low-lying vegetation (e.g. grass) like Ireland – the study area for this investigation. Here, we test the surface scattering mechanism, derived from H-alpha dual-pol decomposition, together with other covariates, to estimate percentages of sand, silt, and clay, over vegetated terrain, using Sentinel 1 data (dual-pol C-band SAR). The statistical modelling approaches evaluated – linear regression (LRM) and tree-based regression models (machine learning) – explicitly consider the compositional nature of soil texture. When compared to the models fitted without surface scattering data, results showed that the inclusion of the surface scattering data improved estimates of silt and clay, with the compositional linear regression model, and estimates of sand and silt fractions with different tree-based models. While not without limitations, our study demonstrated that the polarimetric decomposition method, which is typically used for classification and segmentation purposes, could also be used for soil property estimation, broadening the application of this technique in microwave remote sensing studies.

Identification of aroma compounds responsible for apple flavor via palatability score prediction in tree-based regression models

Apple flavor has been linked to the composition of aroma compounds; however, the impact of varying concentrations of aroma compounds remains poorly understood. While sensory evaluation is commonly used to analyze food palatability, assessing the effect of concentration of all aroma compounds poses challenges. This study screened 174 apple genotypes for significant aroma compounds using regression models, Gas chromatography–mass spectroscopy analysis, and sensory palatability scores. Two models viz. Random Forest and XGBoost were employed to predict palatability scores, surpassing the limitations of correlation analysis. Shapley's additive explanatory value analysis identified 5-hexenol, ethyl 2-methyl butyrate, 2-methylbutyl acetate, (E,E)-α-farnesene, and (Z)-3-hexenol as key aroma compounds. In a sensory evaluation validating the effect of aroma compound concentration, the variation in (E,E)-α-farnesene concentration negatively affected sensory palatability scores. This study demonstrated a novel approach for identifying aroma compounds that impact apple flavor, providing valuable insights to enhance apple quality and palatability.

Forecasting Oil Production of Well 159-F-14H in the Volve Field Using Machine Learning Model

Petroleum engineers require information about the production performance of a well in order to know when the well is no longer feasible to produce. By using the approachment technique of machine learning, the research was conducted using a tree-based regression model, Random Forest Regressor, Extra Trees Regressor, and Gradient Boosting Regressor. This research was done by predicting the production of an existing well in the Volve field, namely well 159-F-14H using its field data; average downhole pressure, average downhole temperature, average wellhead temperature, average wellhead pressure, on-stream hours, average choke size percentage, gas volume from well, water volume from well. The data used is 1093 days and 70% is used for training and as much as 30% for testing. A comparative study was carried out on the predictive performance of the three models. Random Forest shows the best testing result as well as RMSE 5.134 and R2 0.974, followed by Gradient Boosting shows RMSE 5.927 and R2 0.965, and Extra Trees shows RMSE 6.524 and R2 0,958.

Evaluating Familiarity Ratings of Domain Concepts with Interpretable Machine Learning: A Comparative Study

Psycholinguistic properties such as concept familiarity and concreteness have been investigated in relation to technological innovations in teaching and learning. Due to ongoing advances in semantic representation and machine learning technologies, the automatic extrapolation of lexical psycholinguistic properties has received increased attention across a number of disciplines in recent years. However, little attention has been paid to the reliable and interpretable assessment of familiarity ratings for domain concepts. To address this gap, we present a regression model grounded in advanced natural language processing and interpretable machine learning techniques that can predict domain concepts’ familiarity ratings based on their lexical features. Each domain concept is represented at both the orthographic–phonological level and semantic level by means of pretrained word embedding models. Then, we compare the performance of six tree-based regression models (adaptive boosting, gradient boosting, extreme gradient boosting, a light gradient boosting machine, categorical boosting, and a random forest) on domain concepts’ familiarity rating prediction. Experimental results show that categorical boosting with the lowest MAPE (0.09) and the highest R2 value (0.02) is best suited to predicting domain concepts’ familiarity. Experimental results also revealed the prospect of integrating tree-based regression models and interpretable machine learning techniques to expand psycholinguistic resources. Specifically, findings showed that the semantic information of raw words and parts of speech in domain concepts are reliable indicators when predicting familiarity ratings. Our study underlines the importance of leveraging domain concepts’ familiarity ratings; future research should aim to improve familiarity extrapolation methods. Scholars should also investigate the correlation between students’ engagement in online discussions and their familiarity with domain concepts.

Rapid Rice Yield Estimation Using Integrated Remote Sensing and Meteorological Data and Machine Learning

This study developed a rapid rice yield estimation workflow and customized yield prediction model by integrating remote sensing and meteorological data with machine learning (ML). Several issues need to be addressed while developing a crop yield estimation model, including data quality issues, data processing issues, selecting a suitable machine learning model that can learn from few available time-series data, and understanding the non-linear relationship between historical crop yield and remote sensing and meteorological factors. This study applied a series of data processing techniques and a customized ML model to improve the accuracy of crop yield estimation at the district level in Nepal. It was found that remote sensing-derived NDVI product alone was not sufficient for accurate estimation of crop yield. After incorporating other meteorological variables into the ML models, estimation accuracy improved dramatically. Along with NDVI, the meteorological variables of rainfall, soil moisture, and evapotranspiration also exhibited a strong association with rice yield. This study also found that stacking multiple tree-based regression models together could achieve better accuracy than benchmark linear regression or standalone ML models. Due to the unique and distinct physio-geographical setting of each district, a variation in estimation accuracy from district to district could be observed. Our data processing and ML model workflow achieved an average of 92% accuracy of yield estimation with RMSE 328.06 kg/ha and MAE 317.21 kg/ha. This methodological workflow can be replicated in other study areas and the results can help the local authorities and stakeholders understand the factors affecting crop yields as well as estimating crop yield before harvesting season to ensure food security and sustainability.

Weight Prediction for Fishes in Setiu Wetland, Terengganu, using Machine Learning Regression Model

Predicting fish weight holds several essential implications in ecology, such as population assessment, trophic interactions within ecosystems, biodiversity studies of fish communities, ecosystem modelling, habitat evaluation for different fish species, climate change research, and support fisheries management practices. The objective of the studies is to analyse the prediction performance of machine learning (ML) regression models by applying different statistical analysis techniques. This study collected biometric measurements (total length and body weight) for 19 fish families from three locations in Setiu Wetland, Terengganu, captured between 2011 and 2012. The study adopts two regression types: Linear Regression (i.e., Multiple Linear, Lasso, and Ridge model) and Tree-based Regression (i.e., Decision Tree, Random Forest, and XGBoost model). Mean absolute error (MAE), root-mean-square error (RMSE), and coefficient of determination (R2) were used to evaluate performance. The results showed that the proposed ML regression models successfully predicted fish weight in Setiu Wetlands, and the Tree-based Regression model provides more accurate prediction results than the Linear Regression model. As a result, Random Forest is the best predictive model out of the six suggested ML regressions, with the highest accuracy at 96.1% and the lowest RMSE and MAE scores at 3.352 and 0.880, respectively. In conclusion, the use of machine learning is crucial for rapid, precise, and cost-effective fish weight measurement. By incorporating weight prediction into ecological research and management practices, we may make informed decisions supporting the conservation and sustainable use of fish populations and their habitats.

Predicting Steam Games Rating with Regression

This paper tries to find out the best regression model to predict the rating of video games. It is done by comparing multiple variables related to Metascore, such as genres and player count. In order to be able to get accurate results, we gather some data by scraping them from Steam and combining them with public data. The games in this study are from Steam since it is one of the largest computer video games distributors. In this study, we evaluate several regression models, such as Linear regression, Decision Tree, Random Forest to predict the game rating. The experiment shows that tree-based regression model, such as LightGBM and Random Forest performed better than any other regression method, with R2 score above 0.9.

Predictor Selection and Machine Learning Regression Methods to Predict Saturated Hydraulic Conductivity From a Large Public Soil Database

Highlights In this study, six machine learning (ML) models were developed using a large database of soils to predict saturated hydraulic conductivity of these soils using easily measured soil characteristics. Tree-based regression models outperformed all other ML models tested. Neural networks were not suitable for predicting saturated hydraulic conductivity. Clay content, followed by bulk density, explained the highest amount of variation in the data of the predictors examined. Abstract. One of the most important soil hydraulic properties for modeling water transport in the vadose zone is saturated hydraulic conductivity. However, it is challenging to measure it in the field. Pedotransfer Functions (PTFs) are mathematical models that can predict saturated hydraulic conductivity (Ks) from easily measured soil characteristics. Though the development of PTFs for predicting Ks is not new, the tools and methods used to predict Ks are continuously evolving. Model performance depends on choosing soil features that explain the largest amount of Ks variance with the fewest input variables. In addition, the lack of interpretability in most “black box” machine learning models makes it difficult to extract practical knowledge as the machine learning process obfuscates the relationship between inputs and outputs in the PTF models. The objective of this study was to develop a set of new PTFs for predicting Ks using machine learning algorithms and a large database of over 8000 soil samples (the Florida Soil Characterization Database) while incorporating statistical methods to inform predictor selection for the model inputs. Of the machine learning (ML) models tested, random forest regression (RF) and gradient-boosted regression (GB) gave the best performances, with R2 = 0.71 and RMSE = 0.47 cm h-1 on the test data for both. Using the permutation feature importance technique, the GB and RF regression models showed similar results, where clay content described the most variation in the data, followed by bulk density. The implication of this study is that, when predicting Ks using the Florida Soil Characterization Database, priority should be given to obtaining quality data on clay content and bulk density as they are the most influential predictors for estimating Ks. Keywords: Deep learning, Gradient boosted regression, Pedotransfer functions, Random forest regression, Soil database, Soil properties.

Mixed-effect models with trees

Tree-based regression models are a class of statistical models for predicting continuous response variables when the shape of the regression function is unknown. They naturally take into account both non-linearities and interactions. However, they struggle with linear and quasi-linear effects and assume iid data. This article proposes two new algorithms for jointly estimating an interpretable predictive mixed-effect model with two components: a linear part, capturing the main effects, and a non-parametric component consisting of three trees for capturing non-linearities and interactions among individual-level predictors, among cluster-level predictors or cross-level. The first proposed algorithm focuses on prediction. The second one is an extension which implements a post-selection inference strategy to provide valid inference. The performance of the two algorithms is validated via Monte Carlo studies. An application on INVALSI data illustrates the potentiality of the proposed approach.

Machine Learning-Based Risk Stratification for Gestational Diabetes Management.

Gestational diabetes mellitus (GDM) is often diagnosed during the last trimester of pregnancy, leaving only a short timeframe for intervention. However, appropriate assessment, management, and treatment have been shown to reduce the complications of GDM. This study introduces a machine learning-based stratification system for identifying patients at risk of exhibiting high blood glucose levels, based on daily blood glucose measurements and electronic health record (EHR) data from GDM patients. We internally trained and validated our model on a cohort of 1148 pregnancies at Oxford University Hospitals NHS Foundation Trust (OUH), and performed external validation on 709 patients from Royal Berkshire Hospital NHS Foundation Trust (RBH). We trained linear and non-linear tree-based regression models to predict the proportion of high-readings (readings above the UK’s National Institute for Health and Care Excellence [NICE] guideline) a patient may exhibit in upcoming days, and found that XGBoost achieved the highest performance during internal validation (0.021 [CI 0.019–0.023], 0.482 [0.442–0.516], and 0.112 [0.109–0.116], for MSE, R2, MAE, respectively). The model also performed similarly during external validation, suggesting that our method is generalizable across different cohorts of GDM patients.

Abstract 1938: Large-scale identification of mutation activity and drug sensitivity in BRAF via a novel multi-label multi-task CNN model

Abstract Precision medicine has allowed for many drugs to be developed for frequently occurring well studied oncogenic mutations such as NTRK fusions or EGFR exon 19 mutations. However, large scale genomic sequencing of patient samples shows that tumors harbor many mutations classified as Variants of Uncertain Significance (VUS). Such mutations could have a significant role in tumor progression and can thus serve as potential drug targets. Therefore, understanding and characterizing the functional significance of VUSs and their response to targeted agents is essential. Here we present a novel machine learning (ML) model consisting of a multi-label, multi-task deep convolutional neural network followed by a decision tree-based regression model. Focusing on alterations in BRAF, we used data from a cell-based assay that measures the activity of signaling pathway activation. This is done using fluorescent imaging of cells expressing a mutated protein together with a fluorescently labeled signaling pathway reporter, providing the input to our model. We trained the model on 3 types of cell images: cells transfected with WT BRAF, BRAF V600E, and BRAF V600E treated with a high dose of Vemurafenib. We use two datasets to evaluate our performance: a set of 17 known active BRAF fusions, as well as a set of 16 known active non-V600E mutations. We also compare our performance to previously published single-task ML model that aims to detect both activity and response. The two methodologies are compared via two criteria: ability to detect activity of the mutations in the dataset, as well as ability to predict response to Vemurafenib or to FORE8394, a drug previously unseen by the model. We show that while both single-task and multi-task models identify all 17 known active fusions as oncogenic, the multi-task does slightly better on the non-V600E mutations, correctly identifying 15/16 of the active mutations vs 10/16 for the single-task model. Comparing drug response, the multi-task model has higher sensitivity in detecting active mutations as responsive to Vemurafenib or FORE8394, including all V600 mutations which are known responders to Vemurafenib and for whom the single-task model does not capture response. Following the training and validation, given a dataset of >300 previously unseen BRAF mutations, the multi-task model is then able to predict both the mutation’s oncogenicity level as well as its expected response to the given drug. Interestingly, the multi-task model also suggests a different drug response profile for vemurafenib compared to FORE8394. We conclude that our novel multi-task model provides an accurate and efficient method for uncovering the actionable and treatable mutational landscape of a drug for patients with mutations in BRAF. It can thus be viewed as a step forward in developing sensitive methodologies for determining patients that are more susceptible to benefit from a potential drug. Citation Format: Ilona Kifer, Arie Aizenman, Natalie Fillipov-Levy, Zohar Barbash, Michael Vidne, Gabi Tarcic. Large-scale identification of mutation activity and drug sensitivity in BRAF via a novel multi-label multi-task CNN model [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1938.

Adolescent affective psychopathic traits: the long-term outcomes of mother-infant attachment across 14years.

Affective psychopathic traits have been associated with increased violence and aggression. The scholarly interest in the pathways leading to affective psychopathic traits in adolescence has been growing. Infant attachment security may be one of the early life factors that contribute to the development of affective psychopathic traits. In this study, we test the infant attachment styles measured with the Strange Situation Procedure as a predictor of adolescent affective psychopathic traits measured in three dimensions: remorselessness, unemotionality, and callousness, among 1149 families in the Eunice Kennedy Shriver National Institute of Child Health and Human Development Study of Early Child Care and Youth Development (SECCYD). Propensity scores for the four attachment styles were estimated with iterative tree-based regression models. After accounting for the potential confounding effects of demographic characteristics and child temperament with the inverse probability of treatment weighting, weighted generalized linear models revealed the association between insecure/avoidant attachment style measured one year after childbirth and higher levels of remorselessness, unemotionality, and callousness at 15years. These findings provide insights into the long-term outcomes for attachment relationships established in early life.

Interpretation of Machine-Learning-Based (Black-box) Wind Pressure Predictions for Low-Rise Gable-Roofed Buildings Using Shapley Additive Explanations (SHAP)

Conventional methods of estimating pressure coefficients of buildings retain time and cost constraints. Recently, machine learning (ML) has been successfully established to predict wind pressure coefficients. However, regardless of the accuracy, ML models are incompetent in providing end-users’ confidence as a result of the black-box nature of predictions. In this study, we employed tree-based regression models (Decision Tree, XGBoost, Extra-tree, LightGBM) to predict surface-averaged mean pressure coefficient (Cp,mean), fluctuation pressure coefficient (Cp,rms), and peak pressure coefficient (Cp,peak) of low-rise gable-roofed buildings. The accuracy of models was verified using Tokyo Polytechnic University (TPU) wind tunnel data. Subsequently, we used Shapley Additive Explanations (SHAP) to explain the black-box nature of the ML predictions. The comparison revealed that tree-based models are efficient and accurate in wind-predicting pressure coefficients. Interestingly, SHAP provided human-comprehensible explanations for the interaction of variables, the importance of features towards the outcome, and the underlying reasoning behind the predictions. Moreover, SHAP confirmed that tree-based predictions adhere to the flow physics of wind engineering, advancing the fidelity of ML-based predictions.

Explainable Machine Learning (XML) to predict external wind pressure of a low-rise building in urban-like settings

This study used explainable machine learning (XML), a new branch of Machine Learning (ML), to elucidate how ML models make predictions. Three tree-based regression models, Decision Tree (DT), Random Forest (RF), and Extreme Gradient Boost (XGB), were used to predict the normalized mean (Cp,mean), fluctuating (Cp,rms), minimum (Cp,min), and maximum (Cp,max) external wind pressure coefficients of a low-rise building with fixed dimensions in urban-like settings for several wind incidence angles. Two types of XML were used — first, an intrinsic explainable method, which relies on the DT structure to explain the inner workings of the model, and second, SHAP (SHapley Additive exPlanations), a post-hoc explanation technique used particularly for the structurally complex XGB. The intrinsic explainable method proved incapable of explaining the deep tree structure of the DT, but SHAP provided valuable insights by revealing various degrees of positive and negative contributions of certain geometric parameters, the wind incidence angle, and the density of buildings that surround a low-rise building. SHAP also illustrated the relationships between the above factors and wind pressure, and its explanations were in line with what is generally accepted in wind engineering, thus confirming the causality of the ML model's predictions.

Differential impacts of urbanization characteristics on city-level carbon emissions from passenger transport on road: Evidence from 360 cities in China

Although it's well known that the carbon intensity from passenger transport of cities varies widely, few studies assessed the disparities of that in city-level and its underlying factors due to the limited availability of data, and thus developed effective strategies for different types of cities. This study is the first to present a comprehensive inventory of emissions from passenger transport on road for 360 cities in mainland China for 2018, based on the data from 5 transport modes and evaluated by combining distance-based and top-down fuel-based methods. In 2018, passenger transport on road in China emitted 1076 MtC. A large portion of CO2 emissions was identified in the southern and eastern coastal areas and capital cities. GDP, population, and policy were the major factors determining the total CO2 emissions, but not carbon intensity. Clustering analysis of carbon intensity and 9 socio-economic predictors, using a tree-based regression model, clustered the 360 cities into 6 groups and showed that higher carbon intensities occurred in both affluent city groups with a high active population share and less affluent city groups with a low population density but high density of trip destinations. Forward-and-backward stepwise multiple regression analysis indicated that constructing a compact city is more effective for city groups with a high income and high active population share. Enhancing land-use mixed degree is more critical for city groups with a high income and low active population share, while shortening travel distance by intensifying infrastructure construction is more important for the less affluent city groups.