Jianfeng Xi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Jianfeng Xi has a total of 60 co-authors. Here are the top co-authors by number of shared publications:
- Tongqiang Ding
Transportation College, Jilin University, Changchun 130022, China
14 shared publications - Lili Zheng
Transportation College, Jilin University, Changchun 130022, China
6 shared publications - Jian Tian
China Academy of Transportation Sciences, Beijing 100029, China
3 shared publications - Xian-Sheng Li
Jilin University
3 shared publications - Xue-Lian Zheng
Affiliation Transportation College, Jilin University, Changchun, Jilin, China
3 shared publications - Yuan-Yuan Ren
Jilin University
3 shared publications
Latest Publications
- 13 March 2026
- Published by Elsevier in Reliability Engineering & System Safety
- Vol. 272
Interpretable deep learning framework for urban bus traffic safety situation forecast
- 13 March 2026
- Published by Elsevier in Reliability Engineering & System Safety
- Vol. 272
Interpretable deep learning framework for urban bus traffic safety situation forecast
- 12 November 2025
- Published by Taylor & Francis in Traffic Injury Prevention
- Vol. 27  (8) , 1-7
Impact of roadside distributed photovoltaic systems on driver glare
With the widespread application of photovoltaic technology in transportation infrastructure, the potential threat to driving safety posed by glare generated by roadside distributed photovoltaic systems has become a concern. This study aims to evaluate the varying degrees of glare caused by the azimuth(Orient), tilt angle (Tilt), observation point height (Height), and vehicle position on the photovoltaic section (Distance). The Transformer attention mechanism was used to weight the input features, explore the correlation and importance between features, and the extracted features were concatenated with the original features to form a new feature set. The concatenated feature data were used to train Random Forest (RF) and Gradient Boosting Decision Tree (GBDT) models. The RF model after feature processing achieved Coefficient of determination (R2) values of 0.831 and 0.907 on the test sets, respectively, with lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) indicators, showing better overall performance. Therefore, the improved RF model was selected for predicting driver glare duration. It was found that Orient is the most important feature affecting glare, which should be prioritized in the installation of photovoltaic systems. For highway photovoltaic installations, priority should be given to Orient, which can alleviate the situation during periods of high glare. Moreover, Tilt adjustment balances power generation and reduces glare. In addition, the directional offset from the lane at curves or exits further ensures driving safety.
- 12 November 2025
- Published by Taylor & Francis in Traffic Injury Prevention
- Vol. 27  (8) , 1-7
Impact of roadside distributed photovoltaic systems on driver glare
With the widespread application of photovoltaic technology in transportation infrastructure, the potential threat to driving safety posed by glare generated by roadside distributed photovoltaic systems has become a concern. This study aims to evaluate the varying degrees of glare caused by the azimuth(Orient), tilt angle (Tilt), observation point height (Height), and vehicle position on the photovoltaic section (Distance). The Transformer attention mechanism was used to weight the input features, explore the correlation and importance between features, and the extracted features were concatenated with the original features to form a new feature set. The concatenated feature data were used to train Random Forest (RF) and Gradient Boosting Decision Tree (GBDT) models. The RF model after feature processing achieved Coefficient of determination (R2) values of 0.831 and 0.907 on the test sets, respectively, with lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) indicators, showing better overall performance. Therefore, the improved RF model was selected for predicting driver glare duration. It was found that Orient is the most important feature affecting glare, which should be prioritized in the installation of photovoltaic systems. For highway photovoltaic installations, priority should be given to Orient, which can alleviate the situation during periods of high glare. Moreover, Tilt adjustment balances power generation and reduces glare. In addition, the directional offset from the lane at curves or exits further ensures driving safety.
- Open Access
- 14 July 2025
- Published by Springer Nature in Research Square
Identification of Dynamic Driving Styles based on Behavioral Primitives: A Research from Data to Insights
Driving style reflects a driver's vehicle manipulation and driving habits, it is essential for understanding and analyzing a driver’s dynamic behavior and decision-making process. To better understand the intrinsic characteristics of how drivers make decisions and control their vehicles based on external conditions, this paper proposes an unsupervised framework for dynamic driving style recognition based on driving behavior primitives. The framework consists of three key stages: primitive extraction, model development, and dynamic driving style recognition. Unsupervised techniques are employed to extract meaningful driving behavior primitives from long-term driving data. The primitive serves as the fundamental unit for dynamic driving style analysis, and a driving style assessment model is built using a linear weighting approach. This model quantifies the risks of primitives and the transition risks between adjacent primitives. The thresholds for classifying cautious, average, and aggressive driving styles are determined using an improved particle swarm optimization algorithm. The proposed dynamic driving style recognition framework comprehensively considers the characteristics of the current primitive and its surrounding context. By preserving the temporal features of driving behavior, the framework provides fine-grained recognition of dynamic driving styles. Additionally, a deeper understanding of a driver’s dynamic driving style and their long-term driving behavior can be gained based on this framework, which will benefit the development of AVs and ADASs.
- 14 July 2025
- Published by Springer Nature in Research Square
Open AccessIdentification of Dynamic Driving Styles based on Behavioral Primitives: A Research from Data to Insights
Driving style reflects a driver's vehicle manipulation and driving habits, it is essential for understanding and analyzing a driver’s dynamic behavior and decision-making process. To better understand the intrinsic characteristics of how drivers make decisions and control their vehicles based on external conditions, this paper proposes an unsupervised framework for dynamic driving style recognition based on driving behavior primitives. The framework consists of three key stages: primitive extraction, model development, and dynamic driving style recognition. Unsupervised techniques are employed to extract meaningful driving behavior primitives from long-term driving data. The primitive serves as the fundamental unit for dynamic driving style analysis, and a driving style assessment model is built using a linear weighting approach. This model quantifies the risks of primitives and the transition risks between adjacent primitives. The thresholds for classifying cautious, average, and aggressive driving styles are determined using an improved particle swarm optimization algorithm. The proposed dynamic driving style recognition framework comprehensively considers the characteristics of the current primitive and its surrounding context. By preserving the temporal features of driving behavior, the framework provides fine-grained recognition of dynamic driving styles. Additionally, a deeper understanding of a driver’s dynamic driving style and their long-term driving behavior can be gained based on this framework, which will benefit the development of AVs and ADASs.
- Open Access
- 30 April 2025
- Published by Springer Nature in Research Square
Unsupervised inference of driving behavior primitives considering feature correlations and temporal dynamics
As the smallest units of driving behavior, driving behavior primitive plays a crucial role in understanding driving dynamics. To better align the inferred driving behavior primitives with real-world driving scenarios, this paper proposes an unsupervised method consisting of two key stages: data segmentation and segments clustering. In data segmentation, a Hierarchical Bayesian Model-based Agglomerative Sequence Segmentation (H-BMASS) model is introduced to accurately capture the change points of variables across different dimensions, thereby avoiding under-segmentation issue that occurs when BMASS is applied to multivariate time series. In segments clustering, to address the limitation of LDA in capturing temporal dynamics in time series data, an Integrating Distribution- and Trend-based Latent Dirichlet Allocation (IDT-LDA) model is proposed. This model incorporates the change trend of data points to capture the temporal dynamics, determined by the derivatives and fluctuations. Compared to conventional BMASS, H-BMASS effectively identifies the change points of driving behavior, with clear structural differences between adjacent segments. Compared to the primitives obtained using Distribution-based LDA, the primitives clustered using IDT-LDA exhibit improved inter-class separability and enhanced interpretability. This unsupervised framework offers an effective method for inferring driving behavior primitives, which will benefit the development of AVs and ADASs.
- 30 April 2025
- Published by Springer Nature in Research Square
Open AccessUnsupervised inference of driving behavior primitives considering feature correlations and temporal dynamics
As the smallest units of driving behavior, driving behavior primitive plays a crucial role in understanding driving dynamics. To better align the inferred driving behavior primitives with real-world driving scenarios, this paper proposes an unsupervised method consisting of two key stages: data segmentation and segments clustering. In data segmentation, a Hierarchical Bayesian Model-based Agglomerative Sequence Segmentation (H-BMASS) model is introduced to accurately capture the change points of variables across different dimensions, thereby avoiding under-segmentation issue that occurs when BMASS is applied to multivariate time series. In segments clustering, to address the limitation of LDA in capturing temporal dynamics in time series data, an Integrating Distribution- and Trend-based Latent Dirichlet Allocation (IDT-LDA) model is proposed. This model incorporates the change trend of data points to capture the temporal dynamics, determined by the derivatives and fluctuations. Compared to conventional BMASS, H-BMASS effectively identifies the change points of driving behavior, with clear structural differences between adjacent segments. Compared to the primitives obtained using Distribution-based LDA, the primitives clustered using IDT-LDA exhibit improved inter-class separability and enhanced interpretability. This unsupervised framework offers an effective method for inferring driving behavior primitives, which will benefit the development of AVs and ADASs.
- Open Access
- 23 January 2025
- Published by MDPI AG in Sustainability
- Vol. 17  (3) , 906
Modeling of Tank Vehicle Rollover Risk Assessment on Curved–Slope Combination Sections for Sustainable Transportation Safety
Tank vehicles are highly prone to rollover accidents, especially on curved–slope combination sections, which can cause hazardous chemical spills, endangering the environment, public safety, and human health. Therefore, it is crucial to conduct research aimed at reducing the risk of such incidents. Method: The rollover risk of tank vehicles under various loading conditions while traveling on curved–slope combination sections was investigated using driver–vehicle–road dynamics simulation. A multiple linear regression model was then developed to further quantify the impact of key factors on the rollover risk. Results: The results revealed that the road curve radius, vehicle operating speed, and liquid cargo fill level have the greatest impact on a tank vehicle’s rollover risk, and higher fill levels, higher speeds, and steeper downhill slopes all amplify the impact of curve radius on the rollover risk. In some cases, adhering to the road’s speed limit alone was insufficient to ensure the safe passage of the tank vehicle through curves. Conclusions: This study introduced, for the first time, a rollover risk assessment model for tank vehicles operating on curved–slope combination sections. The findings reveal effective methods to improve the transportation safety of tank vehicles. Practical Applications: The findings of this study can assist transportation agencies in selecting routes with lower rollover risks for tank vehicles with different configurations, as well as guide the development of loading standards and curve speed limits. This will effectively reduce rollover accidents of tank vehicles and support sustainable, safer transportation practices.
- 23 January 2025
- Published by MDPI AG in Sustainability
- Vol. 17  (3) , 906
Open AccessModeling of Tank Vehicle Rollover Risk Assessment on Curved–Slope Combination Sections for Sustainable Transportation Safety
Tank vehicles are highly prone to rollover accidents, especially on curved–slope combination sections, which can cause hazardous chemical spills, endangering the environment, public safety, and human health. Therefore, it is crucial to conduct research aimed at reducing the risk of such incidents. Method: The rollover risk of tank vehicles under various loading conditions while traveling on curved–slope combination sections was investigated using driver–vehicle–road dynamics simulation. A multiple linear regression model was then developed to further quantify the impact of key factors on the rollover risk. Results: The results revealed that the road curve radius, vehicle operating speed, and liquid cargo fill level have the greatest impact on a tank vehicle’s rollover risk, and higher fill levels, higher speeds, and steeper downhill slopes all amplify the impact of curve radius on the rollover risk. In some cases, adhering to the road’s speed limit alone was insufficient to ensure the safe passage of the tank vehicle through curves. Conclusions: This study introduced, for the first time, a rollover risk assessment model for tank vehicles operating on curved–slope combination sections. The findings reveal effective methods to improve the transportation safety of tank vehicles. Practical Applications: The findings of this study can assist transportation agencies in selecting routes with lower rollover risks for tank vehicles with different configurations, as well as guide the development of loading standards and curve speed limits. This will effectively reduce rollover accidents of tank vehicles and support sustainable, safer transportation practices.
- Open Access
- 15 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 171348-171367
Analysis of Risk Bus Driver Characteristics and Research on Risk Level Evaluation Methods for Bus Drivers
Currently, there is a lack of a comprehensive and integrated method for assessing risk levels of bus drivers. This study utilizes XGBOOST and Logistic regression models to analyze the impact of various indicator features of bus drivers on crash risks. A grey whitening weight function model is then constructed to evaluate the risk levels of bus drivers, achieving a quantified assessment of their risk levels. Based on the research findings, the following observations were made: (1) The number of non-fault crashes is the most important risk feature influencing the occurrence of at-fault crashes. (2) Features related to crashes, violations, and alarms, as well as age, bus driving experience, driving experience, route length, and the number of stops, have a negative impact on the occurrence of at-fault crashes. (3) The study quantifies bus drivers into five risk levels, with higher levels indicating higher risk. It was found that 94.94% of bus drivers are in the second and third risk levels, 4.93% in the first and fourth risk levels, and only 0.12% of bus drivers are in the highest fifth risk level. The conclusions drawn in this study, along with the proposed method for evaluating risk levels of bus drivers, will contribute to the evaluation and management of bus drivers by bus companies and transportation authorities, thereby reducing crashes in public transportation.
- 15 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 171348-171367
Open AccessAnalysis of Risk Bus Driver Characteristics and Research on Risk Level Evaluation Methods for Bus Drivers
Currently, there is a lack of a comprehensive and integrated method for assessing risk levels of bus drivers. This study utilizes XGBOOST and Logistic regression models to analyze the impact of various indicator features of bus drivers on crash risks. A grey whitening weight function model is then constructed to evaluate the risk levels of bus drivers, achieving a quantified assessment of their risk levels. Based on the research findings, the following observations were made: (1) The number of non-fault crashes is the most important risk feature influencing the occurrence of at-fault crashes. (2) Features related to crashes, violations, and alarms, as well as age, bus driving experience, driving experience, route length, and the number of stops, have a negative impact on the occurrence of at-fault crashes. (3) The study quantifies bus drivers into five risk levels, with higher levels indicating higher risk. It was found that 94.94% of bus drivers are in the second and third risk levels, 4.93% in the first and fourth risk levels, and only 0.12% of bus drivers are in the highest fifth risk level. The conclusions drawn in this study, along with the proposed method for evaluating risk levels of bus drivers, will contribute to the evaluation and management of bus drivers by bus companies and transportation authorities, thereby reducing crashes in public transportation.
- Open Access
- 8 October 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (19) , 9094
Grading of Traffic Interruptions in Highways to Tibet Based on the Entropy Weight-TOPSIS Method and Fuzzy C-Means Clustering Algorithm
The interruption of transportation on the way to Tibet has brought great losses to the Tibetan region. The work proposed a model that integrated the entropy weight-TOPSIS method with the fuzzy C-means clustering algorithm to discuss the causes and characteristics of traffic interruptions on the four main highways to Tibet. This approach aimed to quantify and grade traffic interruption states. First, the entropy weight-TOPSIS method was used to mitigate dimensions among various indices and quantitatively evaluate the status values of traffic interruptions. Then, the fuzzy C-means clustering algorithm was employed to grade these values. The proposed model graded traffic interruption states into four levels by evaluating the duration, mileage, and severity of traffic interruptions. Moreover, the four-level classification scheme can reflect the severity of traffic blocking events more precisely while maintaining a lower PE (Partition Entropy) value. In the four-level classification, the Sichuan–Tibet Highway and Xinjiang–Tibet Highway experienced more level-3 and level-4 serious interruptions, while most high-level interruptions on the Qinghai–Tibet Highway were classified as level-2 ordinary interruptions. The Yunnan–Tibet Highway, with limited data and primarily level-1 classification, was not analyzed in detail. These findings provide a reference for highway management departments to formulate targeted maintenance and emergency measures, especially the Sichuan–Tibet highway, which needs more attention and resource investment to improve its disaster resistance and reduce the impact of traffic interruptions.
- 8 October 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (19) , 9094
Open AccessGrading of Traffic Interruptions in Highways to Tibet Based on the Entropy Weight-TOPSIS Method and Fuzzy C-Means Clustering Algorithm
The interruption of transportation on the way to Tibet has brought great losses to the Tibetan region. The work proposed a model that integrated the entropy weight-TOPSIS method with the fuzzy C-means clustering algorithm to discuss the causes and characteristics of traffic interruptions on the four main highways to Tibet. This approach aimed to quantify and grade traffic interruption states. First, the entropy weight-TOPSIS method was used to mitigate dimensions among various indices and quantitatively evaluate the status values of traffic interruptions. Then, the fuzzy C-means clustering algorithm was employed to grade these values. The proposed model graded traffic interruption states into four levels by evaluating the duration, mileage, and severity of traffic interruptions. Moreover, the four-level classification scheme can reflect the severity of traffic blocking events more precisely while maintaining a lower PE (Partition Entropy) value. In the four-level classification, the Sichuan–Tibet Highway and Xinjiang–Tibet Highway experienced more level-3 and level-4 serious interruptions, while most high-level interruptions on the Qinghai–Tibet Highway were classified as level-2 ordinary interruptions. The Yunnan–Tibet Highway, with limited data and primarily level-1 classification, was not analyzed in detail. These findings provide a reference for highway management departments to formulate targeted maintenance and emergency measures, especially the Sichuan–Tibet highway, which needs more attention and resource investment to improve its disaster resistance and reduce the impact of traffic interruptions.
- Open Access
- 3 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 118533-118550
A Multimodal Driver Anger Recognition Method Based on Context-Awareness
In today’s society, the harm of driving anger to traffic safety is increasingly prominent. With the development of human-computer interaction and intelligent transportation systems, the application of biometric technology in driver emotion recognition has attracted widespread attention. This study proposes a context-aware multi-modal driver anger emotion recognition method (CA-MDER) to address the main issues encountered in multi-modal emotion recognition tasks. These include individual differences among drivers, variability in emotional expression across different driving scenarios, and the inability to capture driving behavior information that represents vehicle-to-vehicle interaction. The method employs Attention Mechanism-Depthwise Separable Convolutional Neural Networks (AM-DSCNN), an improved Support Vector Machines (SVM), and Random Forest (RF) models to perform multi-modal anger emotion recognition using facial, vocal, and driving state information. It also uses Context-Aware Reinforcement Learning (CA-RL) based adaptive weight distribution for multi-modal decision-level fusion. The results show that the proposed method performs well in emotion classification metrics, with an accuracy and F1 score of 91.68% and 90.37%, respectively, demonstrating robust multi-modal emotion recognition performance and powerful emotion recognition capabilities.
- 3 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 118533-118550
Open AccessA Multimodal Driver Anger Recognition Method Based on Context-Awareness
In today’s society, the harm of driving anger to traffic safety is increasingly prominent. With the development of human-computer interaction and intelligent transportation systems, the application of biometric technology in driver emotion recognition has attracted widespread attention. This study proposes a context-aware multi-modal driver anger emotion recognition method (CA-MDER) to address the main issues encountered in multi-modal emotion recognition tasks. These include individual differences among drivers, variability in emotional expression across different driving scenarios, and the inability to capture driving behavior information that represents vehicle-to-vehicle interaction. The method employs Attention Mechanism-Depthwise Separable Convolutional Neural Networks (AM-DSCNN), an improved Support Vector Machines (SVM), and Random Forest (RF) models to perform multi-modal anger emotion recognition using facial, vocal, and driving state information. It also uses Context-Aware Reinforcement Learning (CA-RL) based adaptive weight distribution for multi-modal decision-level fusion. The results show that the proposed method performs well in emotion classification metrics, with an accuracy and F1 score of 91.68% and 90.37%, respectively, demonstrating robust multi-modal emotion recognition performance and powerful emotion recognition capabilities.
- Open Access
- 28 December 2023
- Published by MDPI AG in Applied Sciences
- Vol. 14  (1) , 279
Accident Probability Prediction and Analysis of Bus Drivers Based on Occupational Characteristics
A city bus carries a large number of passengers, and any traffic accidents can lead to severe casualties and property losses. Hence, predicting the likelihood of accidents among bus drivers is paramount. This paper considered occupational driving characteristics such as cumulative driving duration, station entry and exit features, and peak driving times, and categorical boosting (CatBoost) was used to construct an accident probability prediction model. Its effectiveness was confirmed by the daily management data of a Chongqing bus company in June. For data processing, Multiple Imputation by Chained Equations for Random Forests (MICEForest) was used for data filling. In terms of prediction, a comparative analysis of four boosted trees revealed that CatBoost exhibited superior performance. To analyze the critical factors affecting the probability of bus driver accidents, SHapley Additive exPlanations (SHAP) was applied to visualize and interpret the results. In addition to the significant effects of age, rainfall, and azimuthal change, etc., we innovatively discovered that the proportion of driving duration during peak duration, the dispersion when entering and exiting stations, the proportion of driving duration within a week, and the accumulated driving duration of the previous week also had varying degrees of impact on accident probability. Our research and findings provide a new idea of accident prediction for professional drivers and direct theoretical support for the accident risk management of bus drivers.
- 28 December 2023
- Published by MDPI AG in Applied Sciences
- Vol. 14  (1) , 279
Open AccessAccident Probability Prediction and Analysis of Bus Drivers Based on Occupational Characteristics
A city bus carries a large number of passengers, and any traffic accidents can lead to severe casualties and property losses. Hence, predicting the likelihood of accidents among bus drivers is paramount. This paper considered occupational driving characteristics such as cumulative driving duration, station entry and exit features, and peak driving times, and categorical boosting (CatBoost) was used to construct an accident probability prediction model. Its effectiveness was confirmed by the daily management data of a Chongqing bus company in June. For data processing, Multiple Imputation by Chained Equations for Random Forests (MICEForest) was used for data filling. In terms of prediction, a comparative analysis of four boosted trees revealed that CatBoost exhibited superior performance. To analyze the critical factors affecting the probability of bus driver accidents, SHapley Additive exPlanations (SHAP) was applied to visualize and interpret the results. In addition to the significant effects of age, rainfall, and azimuthal change, etc., we innovatively discovered that the proportion of driving duration during peak duration, the dispersion when entering and exiting stations, the proportion of driving duration within a week, and the accumulated driving duration of the previous week also had varying degrees of impact on accident probability. Our research and findings provide a new idea of accident prediction for professional drivers and direct theoretical support for the accident risk management of bus drivers.
- Open Access
- 14 February 2023
- Published by MDPI AG in Sustainability
- Vol. 15  (4) , 3520
Bus Fleet Accident Prediction Based on Violation Data: Considering the Binding Nature of Safety Violations and Service Violations
The number and severity of bus traffic accidents are increasing annually. Therefore, this paper uses the historical data of Chongqing Liangjiang Public Transportation Co., Ltd. bus driver safety violations, service violations, and road traffic accidents from January to June 2022 and constructs road traffic accident prediction models using Extra Trees, BP Neural Network, Support Vector Machine, Gradient Boosting Tree, and XGBoost. The effects of safety and service violations on vehicular accidents are investigated. The quality of the prediction models is measured by five indicators: goodness of fit, mean square error, root mean square error, mean absolute error, and mean absolute percentage error. The results indicate that the XGBoost model provides the most accurate predictions. Additionally, simultaneously considering safety and service violations can improve the accuracy of the model’s predictions compared to a model that only considers safety violations. Bus safety violations, bus service violations, and bus safety operation violations significantly influence traffic accidents, which account for 27.9%, 20%, and 16.5%, respectively. In addition to safety violations, the service violation systems established by bus companies, such as bus service codes, can be an effective method of regulating the behavior of bus drivers and reducing accidents. They are improving both the safety and quality of public transportation.
- 14 February 2023
- Published by MDPI AG in Sustainability
- Vol. 15  (4) , 3520
Open AccessBus Fleet Accident Prediction Based on Violation Data: Considering the Binding Nature of Safety Violations and Service Violations
The number and severity of bus traffic accidents are increasing annually. Therefore, this paper uses the historical data of Chongqing Liangjiang Public Transportation Co., Ltd. bus driver safety violations, service violations, and road traffic accidents from January to June 2022 and constructs road traffic accident prediction models using Extra Trees, BP Neural Network, Support Vector Machine, Gradient Boosting Tree, and XGBoost. The effects of safety and service violations on vehicular accidents are investigated. The quality of the prediction models is measured by five indicators: goodness of fit, mean square error, root mean square error, mean absolute error, and mean absolute percentage error. The results indicate that the XGBoost model provides the most accurate predictions. Additionally, simultaneously considering safety and service violations can improve the accuracy of the model’s predictions compared to a model that only considers safety violations. Bus safety violations, bus service violations, and bus safety operation violations significantly influence traffic accidents, which account for 27.9%, 20%, and 16.5%, respectively. In addition to safety violations, the service violation systems established by bus companies, such as bus service codes, can be an effective method of regulating the behavior of bus drivers and reducing accidents. They are improving both the safety and quality of public transportation.