A traffic accident loss prediction method based on machine learning and indicator screening

Abstract

The present invention discloses a method for predicting traffic accident losses based on machine learning and index screening, and belongs to the technical field of traffic accident prediction; the present invention constructs a traffic accident loss prediction model based on extra-trees, and predicts the traffic accident loss value through model construction and data training; the prediction accuracy is significantly improved by using an innovative index screening method; the prediction accuracy interval of the accident loss is found through multiple prediction verifications, the stability of the prediction results is guaranteed, and the prediction performance is further improved on the basis of existing prediction research; the present invention uses an index dimensionality reduction method to perform two types of index screening, and improves the prediction accuracy in the process of dimensionality reduction. By comparing the results of different screening methods, multiple influencing indicators are gradually determined based on the prediction accuracy; an index importance evaluation system is established, and important indicators in traffic accidents are obtained by quantitative calculation from multiple angles; targeted measures and suggestions are proposed to provide a reference for the safe development of the industry.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic accident prediction, and in particular to a traffic accident loss prediction method based on machine learning and indicator screening. Background Art

[0002] The research on traffic accident prediction can be roughly divided into two categories. The first is to use traditional prediction methods or machine learning methods to improve the existing prediction methods through various optimization methods, such as parameter adjustment and model combination, in order to obtain better prediction results and improve prediction accuracy. The second is to investigate and analyze the mechanism and causes of traffic accidents, starting from various influencing factors, sort out the relevant indicators of traffic accidents, and explore the correlation between the factors affecting traffic accidents and accidents.

[0003] Both prediction research ideas can effectively predict traffic accident losses. In the research, prediction research and accident influencing factors should be combined to improve prediction efficiency and indicator selection, thus laying a good foundation for scientific prediction. At present, the research in the field of traffic accident loss prediction mainly has the following problems:

[0004] (1) Incomplete data. Traffic accident loss prediction methods usually rely on historical traffic accident data. The comprehensiveness and accuracy of the data are very important for prediction research. The current research on traffic accident losses is limited by the limited data sources and missing statistical data, and the data is not complete enough. When predicting accident losses, the data indicators are often relatively simple and cannot fully explain the severity of the accident losses.

[0005] (2) The applicability of traffic accident loss models is poor. Among the numerous accident prediction models, traditional prediction models generally rely on overall historical data for prediction, while ignoring the specific influencing factors of the accident, such as road conditions and transportation-related factors. These characteristic factors related to traffic accident losses are also of great significance to the accurate prediction of the model. In recent years, machine learning prediction models have been widely used in the field of traffic accident loss prediction, such as neural network models. Due to their strong nonlinear mapping capabilities for traffic accident data, they are widely used to continuously adjust and approach the optimal results through the training process of a large amount of data. In the prediction process, selecting an appropriate prediction model is crucial to improving the prediction accuracy and ensuring the comprehensiveness of the prediction.

[0006] (3) The prediction lacks interpretability. The causes of traffic accidents are complex and there are many influencing factors. The purpose of predicting traffic accident losses is to provide reference and help for accident prevention. The current prediction research lacks interpretability of the prediction results. Some macro predictions of traffic accident losses have not yet introduced exogenous variables, so that it is impossible to provide targeted guidance on the existing traffic safety situation through the prediction results. The practical significance of the prediction needs further study. Although another part of the research has introduced variables related to the transportation industry or accident losses, the impact of various relevant influencing factors on the prediction of traffic accident losses has been taken into account in the prediction, but the potential relationship between each indicator and the prediction target has not been analyzed and further explored, and the interpretability of the indicator variables is still insufficient. Therefore, it is very necessary to include the discussion of the relationship between indicators and predictions in the research on traffic accident losses, which will help improve the effectiveness of accident prevention decisions.

[0007] In order to make more accurate predictions of traffic accidents and improve the prediction effect, existing technologies have proposed some prediction schemes. By constructing prediction models based on different theories, the characteristics of factors affecting traffic accidents are deeply explored to improve the prediction accuracy and enhance the robustness of prediction. For example, the "A Traffic Accident Prediction Method" disclosed in the patent application number "CN116110219A" combines the traffic accident prediction method based on system identification and the traffic accident prediction method based on LSTM, and makes prediction comparisons. The prediction results show that the prediction ability of road traffic accidents can be improved by the combined methods used. The present invention has higher prediction accuracy than the existing methods. Although the highway traffic flow data series has been analyzed, trained and predicted, the causal factors and related indicators of traffic accidents have not been discussed in the invention, and there is a lack of guiding significance for accident prevention.

[0008] The patent application number "CN109598929B" discloses "a method for predicting the number of multi-category traffic accidents". The present invention takes into account the inherent correlation of multi-category traffic accident data, deeply explores the characteristics of factors affecting traffic accidents, and makes simultaneous predictions for the number of multi-category traffic accidents, with high prediction accuracy and good robustness. In particular, it has a good prediction effect in predicting the number of major and extremely major traffic accidents, which are of great concern in the field of traffic safety. It is a relatively comprehensive and hierarchical method for predicting the number of traffic accidents, which can meet the needs of traffic accident prediction data in the fields of traffic safety hazard identification, accident black spot investigation, road improvement design, etc. However, the analysis of the independent variables of traffic accidents in the invention does not involve the comparison of the importance of indicators. Therefore, the research on the factors affecting accidents still remains at the qualitative level, and lacks practical significance at the macro level of prediction.

[0009] Patent application number "CN109034264A" discloses "A CSP-CNN model for predicting the severity of traffic accidents and its modeling method". This invention uses data mining technology to model and predict the severity of traffic accidents based on deep learning. In the invention, although the severity of accident casualties is subdivided into minor, serious, and fatal traffic accidents, the influencing factors and important accident impact indicators are still not analyzed and quantitatively compared, which lacks innovation.

[0010] At present, in the field of traffic accident loss prediction, a major direction is to combine the advantages of multiple models for combined prediction, which helps to improve the accuracy of the prediction and the comprehensiveness of the prediction. In addition, combining qualitative and quantitative prediction methods with computer technology, deep learning algorithms, etc. is also a trend in prediction research. Although the above patents have established prediction models and effectively improved the accuracy and precision of predictions, they have not explored the impact on accident losses at the indicator level, and there are still certain limitations in the research on the practical significance of accident prediction. In view of this, the present invention proposes a traffic accident loss prediction method based on machine learning and indicator screening. Summary of the invention

[0011] The purpose of the present invention is to propose a traffic accident loss prediction method based on machine learning and index screening to solve the problem of low prediction accuracy of the existing accident prediction model in the traffic field. The present invention can effectively improve the prediction accuracy of traffic accidents by using machine learning models, parameter adjustment and model optimization methods, and provide more accurate prediction results for accident prevention.

[0012] In order to achieve the above object, the present invention adopts the following technical solutions:

[0013] A traffic accident loss prediction method based on machine learning and indicator screening includes the following contents:

[0014] S1. Data preprocessing: Collect traffic accident data publicly released on the Internet, preprocess the collected original data on traffic accident losses, and use it as the basis for subsequent data analysis;

[0015] S2. Data classification: the pre-processed data is classified into two categories: input data and output data, wherein the input data includes indicator data related to the transportation industry, and the output data includes prediction data of accident losses;

[0016] S3. Constructing an accident loss prediction model: constructing a traffic accident loss prediction model based on an extremely random tree, and using the constructed model to predict the loss values ​​of various traffic accidents;

[0017] S4. Indicator screening: Construct an importance quantitative evaluation method based on indicator screening, determine two indicator screening methods: a screening method based on average importance and a screening method based on importance ratio, and select the best indicator set based on the above two indicator screening methods based on prediction accuracy;

[0018] S5, global correlation analysis of indicators: calculate the global correlation of multiple influencing indicators in the best indicator set obtained in S4, and rank them;

[0019] S6. Analysis of global fitting effect of indicators: Calculate the global fitting effect of multiple influencing indicators in the best indicator set obtained in S4 and rank them;

[0020] S7, indicator global importance analysis: calculate the global importance of multiple influencing indicators in the best indicator set obtained in S4, and rank them;

[0021] S8. Determine important indicators: Based on the global correlation, global fitting effect, global importance and ranking of the characteristic indicators obtained in S5-S7, each indicator is calculated using the dynamic weighting method to determine the comprehensive importance score of the indicator, and the most important indicator is obtained through analysis and comparison.

[0022] Preferably, the prediction target of the traffic accident loss prediction model based on the extreme random tree is the traffic accident loss value, and the characteristic indicators of the model include relevant influencing factors of the transportation industry and macroeconomic indicators of the national economy; the traffic accident loss value specifically includes the number of accidents, the number of deaths, the number of injuries and property losses.

[0023] Preferably, the screening method based on average importance specifically refers to: calculating the average importance of the indicator set according to the importance of the indicator, retaining those with higher importance than the average, and eliminating those with lower importance than the average;

[0024] The screening method based on importance ratio specifically refers to: sorting each indicator according to the value of importance, retaining the indicators with importance in the top 80%, and eliminating the remaining indicators.

[0025] Preferably, the importance of an indicator is measured based on the idea of ​​average accuracy decrease, random noise is added to a certain indicator, and the importance of the indicator is measured by calculating the decrease in accuracy.

[0026] Preferably, the S5 specifically includes the following contents:

[0027] The linear correlation between two variables is measured by the Pearson correlation coefficient, and the value of the Pearson correlation coefficient is between -1 and 1. After determining the correlation between multiple influencing indicators and different accident losses, the global correlation is calculated: the overall correlation coefficient is recorded as ρ, and the Pearson correlation coefficient is recorded as r. The specific function is expressed as:

[0028]

[0029] Among them, ρ is the overall correlation coefficient, which indicates the overall linear correlation between the two variables in all samples. r is the Pearson correlation coefficient, which indicates the linear correlation between the two variables. The value range is between -1 and 1. r = 1 indicates a perfect positive correlation, r = -1 indicates a perfect negative correlation, and r = 0 indicates no linear correlation. σ X and σ Y Represents the standard deviation of variable X and variable Y, μ X and μ Y Represents the mean of variable X and variable Y respectively. i and Y i are the i-th observation values ​​of the two variables, and are the average values ​​of the two variables.

[0030] The Pearson correlation coefficient is used to analyze the correlation between characteristic indicators and accident losses, determine the correlation between characteristic indicators with multiple influences and accident losses and the size of the global correlation, and then derive the ranking of the global correlation among the multiple influencing factors of traffic accidents and determine the indicators with the closest correlation.

[0031] Preferably, the S6 specifically includes the following contents:

[0032] The square residual coefficient of the regression model is used to measure the degree of fit of the regression model. The value of the square residual coefficient is between 0 and 1. The closer the square residual coefficient is to 1, the stronger the explanatory power of the independent variable on the output variable, and the better the degree of fit of the regression model. Conversely, the worse the degree of fit of the regression model is. The specific function is expressed as:

[0033] R 2 =1-(SSR / SST)

[0034] Among them, R 2 represents the square residual coefficient; SSR is the regression sum of squares, which represents the sum of squares of the difference between the model predicted value and the actual value; SST is the total sum of squares, which represents the sum of squares of the difference between the actual value and the mean;

[0035] By calculating the global regression effects of characteristic indicators and accident losses through regression analysis, the ranking of the global regression effects among the multiple influencing factors of traffic accidents is obtained, and the indicators with the best regression effect are determined.

[0036] Preferably, the S7 specifically includes the following contents:

[0037] The weight of the number of indicators is introduced to calculate the importance of characteristic indicators to different accident indicators. The specific function is expressed as:

[0038]

[0039] Among them, importance(F) represents the global importance of the indicator; importance(f) represents the importance of the characteristic indicator corresponding to the accident indicator; N i Indicates the number of metrics in the metrics collection.

[0040] Compared with the prior art, the present invention provides a traffic accident loss prediction method based on machine learning and index screening, which has the following beneficial effects:

[0041] (1) The present invention adopts two index screening methods: a screening method based on average importance and a screening method based on importance ratio. The index screening and model prediction are then performed, which effectively improves the prediction accuracy of traffic accident losses. By analyzing and comparing the prediction effects under different screening methods, the optimal prediction index set is found.

[0042] (2) The present invention proposes an evaluation method for the importance of evaluation indicators based on a quantitative method, comprehensively considering the importance, fitting effect and correlation of the indicators in the prediction process, and adopts a dynamic weighting method to calculate the importance of each multi-influencing indicator, thereby finding the important indicators in the prediction process and providing a reference for accident prediction and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of calculating the importance mentioned in Example 1 of the present invention;

[0044] Figure 2 This is a flowchart of the quantitative calculation of the indicator importance mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0046] Due to the complex mechanism of traffic accident causes, it is of great significance to consider the influence of exogenous variables related to traffic accidents. In the prediction process, different prediction models have different applicability and advantages, and it is necessary to select a suitable model for prediction. The present invention proposes a traffic accident loss prediction method based on machine learning and index screening. Through the Extra-Tree model, the traffic accident loss value is predicted through model construction and data training. The prediction accuracy is significantly improved by using an innovative index screening method. The prediction accuracy interval of the accident loss is found through multiple prediction verifications, which effectively ensures the stability of the prediction results. The prediction performance is further improved on the basis of other prediction studies. Based on the theory of the Extremely Random Tree, the index dimension reduction method is used to screen the indexes in two ways, and the prediction accuracy is improved in the process of dimension reduction. By comparing the results of different screening methods, multiple influencing indicators are gradually determined based on the prediction accuracy. An index importance evaluation system is established, and important indicators in traffic accidents are quantitatively calculated from multiple angles. Targeted measures and suggestions are proposed to provide a reference for the safe development of the industry.

[0047] The traffic accident loss prediction method based on machine learning and indicator screening proposed by the present invention is described below with reference to specific drawings and example data, which specifically includes the following contents.

[0048] Embodiment 1:

[0049] The present invention proposes a traffic accident loss prediction method based on machine learning and index screening, which includes the following contents:

[0050] Step 1: Data preprocessing:

[0051] Collect traffic accident data publicly released online, pre-process the collected raw data on traffic accident losses, and use it as the basis for subsequent data analysis;

[0052] Step 2: Data classification:

[0053] Dividing the preprocessed data into two categories: input data and output data, wherein the input data includes indicator data related to the transportation industry, and the output data includes forecast data of accident losses;

[0054] Step 3: Build an accident loss prediction model:

[0055] An Extra-Tree traffic accident loss prediction model is constructed. The prediction target is the traffic accident loss value, including the number of accidents, number of deaths, number of injuries and property losses. The characteristic indicators include relevant influencing factors of the transportation industry and macroeconomic indicators, etc., which can be selected according to the actual prediction situation. After the model is established, each loss value is predicted separately.

[0056] Step 4: Indicator screening:

[0057] The extra-tree model constructed in step three is used to predict traffic accident losses. First, the input indicators are screened. Based on the original indicator set, two indicator screening and dimensionality reduction methods are used, namely, the screening method based on average importance and the screening method based on importance ratio.

[0058] Screening method 1 (indicator screening method based on average importance): According to the importance of the indicators, the average importance of the indicator set is calculated, those with higher importance than the average are retained, and those with lower importance than the average are eliminated.

[0059] Screening method 2 (screening based on the top 80% of importance): sort each indicator according to its importance value, retain the indicators with the top 80% of importance, and eliminate the rest.

[0060] In order to measure the feature importance of indicators related to traffic accident losses, we can use the idea of ​​average accuracy decline, that is, add random noise to a certain indicator, and measure the importance of the indicator by calculating the decline in accuracy. If an indicator has a great influence on the result of the prediction task, it can be considered to have a high importance. The calculation process of importance is as follows: Figure 1 shown.

[0061] Step 5: Quantitative analysis of indicator correlation:

[0062] The Pearson correlation coefficient is used to measure the linear correlation between two variables, and its value is between -1 and 1. After determining the correlation between multiple influencing indicators and different accident losses, the global correlation is calculated. The overall correlation coefficient is denoted by ρ, and the Pearson correlation coefficient is denoted by r:

[0063]

[0064]

[0065] Among them, ρ is the overall correlation coefficient, which indicates the overall linear correlation between the two variables in all samples. r is the Pearson correlation coefficient, which indicates the linear correlation between the two variables. The value range is between -1 and 1. r = 1 indicates a perfect positive correlation, r = -1 indicates a perfect negative correlation, and r = 0 indicates no linear correlation. σ X and σ Y Represents the standard deviation of variable X and variable Y, μ X and μ Y Represents the mean of variable X and variable Y respectively. i and Y i are the i-th observation values ​​of the two variables, and are the average values ​​of the two variables.

[0066] The Pearson correlation coefficient is used to analyze the correlation between characteristic indicators and accident losses, so that the correlation between characteristic indicators with multiple influences and accident losses and the size of the global correlation can be determined. Therefore, the ranking of the global correlation among the multiple influencing factors of traffic accidents can be obtained, and the indicator with the closest correlation can be determined.

[0067] Step 6: Analysis of indicator regression effect:

[0068] R for regression models 2 It is a statistic that measures the degree of fit of the regression model, with a value between 0 and 1. The closer it is to 1, the stronger the ability of the independent variable to explain the output variable, and the better the regression effect. Conversely, the regression model effect is worse. The calculation formula is:

[0069] R 2 =1-(SSR / SST)

[0070] Among them, SSR stands for regression sum of squares, which is the sum of squares of the difference between the model prediction value and the actual value. SST is the total sum of squares, which refers to the sum of squares of the difference between the actual value and the mean. Generally speaking, R 2 If the value is greater than 0.8, it indicates that the regression model has a high degree of fit; if it is less than 0.5, it indicates that the regression model has a low degree of fit and the model may need to be adjusted.

[0071] By calculating the global regression effects of characteristic indicators and accident losses through regression analysis, we can rank the global regression effects of multiple influencing factors of traffic accidents and determine the indicators with the best regression effects.

[0072] Step 7: Global importance analysis:

[0073] The weight of the number of indicators is introduced to calculate the importance of characteristic indicators to different accident indicators. Global importance N i is the number of indicators in the set, importance(f) is the importance of the characteristic indicator corresponding to the accident indicator, and importance(F) is the global importance of the indicator.

[0074]

[0075] The importance theory is used to calculate the global importance values ​​of characteristic indicators and accident losses, obtain the global importance ranking of multiple influencing factors of traffic accidents, and determine the indicator with the highest global importance.

[0076] Step 8: Determine important indicators using the dynamic weighting method:

[0077] After the above steps, the global relevance, global regression effect, global importance and ranking of the characteristic indicators are determined. On this basis, each indicator is calculated by dynamic weighting method to determine the comprehensive importance score of the indicator. On the basis of average weighting, the dynamic weight calculation method is added to analyze the indicator factors that have the greatest impact on accident prediction from an objective perspective. The weight of the indicator importance is set to 90%, and the weights of the correlation and goodness of fit are evenly distributed and set to 5% respectively, and the importance of each indicator is calculated. Then reduce the importance weight to 80%, 70%, 60%, and finally 30%, until the weight is reduced to 10%, and the values ​​of each indicator importance, fitting effect and correlation are weighted according to the weight distribution. The result after weight distribution is finally determined to determine the most important indicator.

[0078] The quantitative calculation process of indicator importance is as follows Figure 2 As shown, through the quantitative calculation of the importance of the above indicators, we can eventually determine the indicator with the highest importance to accident losses among the characteristic indicators, that is, the characteristic factor most closely related to the occurrence of traffic accidents, which can provide a reference for macro-forecasting and related policy formulation.

[0079] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A traffic accident loss prediction method based on machine learning and index screening, characterized in that: It includes the following: S1. Data preprocessing: Collect traffic accident data publicly released on the Internet, preprocess the collected original data on traffic accident losses, and use it as the basis for subsequent data analysis; S2. Data classification: the pre-processed data is classified into two categories: input data and output data, wherein the input data includes indicator data related to the transportation industry, and the output data includes prediction data of accident losses; S3. Constructing an accident loss prediction model: constructing a traffic accident loss prediction model based on an extremely random tree, and using the constructed model to predict the loss values ​​of various traffic accidents; S4. Indicator screening: Construct an importance quantitative evaluation method based on indicator screening, determine two indicator screening methods: a screening method based on average importance and a screening method based on importance ratio, and select the best indicator set based on the above two indicator screening methods based on prediction accuracy; S5, global correlation analysis of indicators: calculate the global correlation of multiple influencing indicators in the best indicator set obtained in S4, and rank them; S6. Analysis of global fitting effect of indicators: Calculate the global fitting effect of multiple influencing indicators in the best indicator set obtained in S4 and rank them; S7, indicator global importance analysis: calculate the global importance of multiple influencing indicators in the best indicator set obtained in S4, and rank them; S8. Determine important indicators: Based on the global correlation, global fitting effect, global importance and ranking of the characteristic indicators obtained in S5-S7, each indicator is calculated using the dynamic weighting method to determine the comprehensive importance score of the indicator, and the most important indicator is obtained through analysis and comparison.

2. The method for predicting traffic accident losses based on machine learning and index screening according to claim 1, characterized in that: The prediction target of the traffic accident loss prediction model based on the extreme random tree is the traffic accident loss value. The characteristic indicators of the model include relevant influencing factors of the transportation industry and macroeconomic indicators of the national economy; the traffic accident loss value specifically includes the number of accidents, the number of deaths, the number of injuries and property losses.

3. The method for predicting traffic accident losses based on machine learning and index screening according to claim 1, characterized in that: The screening method based on average importance specifically refers to: according to the importance of the indicators, the average importance of the indicator set is calculated, the indicators with higher importance than the average are retained, and the indicators with lower importance than the average are eliminated; The screening method based on importance ratio specifically refers to: sorting each indicator according to the value of importance, retaining the indicators with importance in the top 80%, and eliminating the remaining indicators.

4. The method for predicting traffic accident losses based on machine learning and index screening according to claim 3 is characterized in that: The importance of indicators is measured based on the idea of ​​average accuracy decline. Random noise is added to an indicator, and the importance of the indicator is measured by calculating the decline in accuracy.

5. The method for predicting traffic accident losses based on machine learning and index screening according to claim 1, characterized in that: The S5 specifically includes the following contents: The linear correlation between two variables is measured by the Pearson correlation coefficient, and the value of the Pearson correlation coefficient is between -1 and 1. After determining the correlation between multiple influencing indicators and different accident losses, the global correlation is calculated: the overall correlation coefficient is recorded as ρ, and the Pearson correlation coefficient is recorded as r. The specific function is expressed as: Among them, ρ is the overall correlation coefficient, which indicates the overall linear correlation between two variables in all samples; r is the Pearson correlation coefficient, which indicates the linear correlation between two variables, and the value range is between -1 and 1, r = 1 indicates a perfect positive correlation, r = -1 indicates a perfect negative correlation, and r = 0 indicates no linear correlation; σ X and σ Y Respectively represent the standard deviation of variable X and variable Y; μ X and μ Y Represents the mean of variable X and variable Y respectively; X i and Y i Respectively represent the i-th observation value of the two variables; and Respectively represent the average values ​​of two variables; The Pearson correlation coefficient is used to analyze the correlation between characteristic indicators and accident losses, determine the correlation between characteristic indicators with multiple influences and accident losses and the size of the global correlation, and then derive the ranking of the global correlation among the multiple influencing factors of traffic accidents and determine the indicators with the closest correlation.

6. The method for predicting traffic accident losses based on machine learning and index screening according to claim 1, characterized in that: The S6 specifically includes the following contents: The square residual coefficient of the regression model is used to measure the degree of fit of the regression model. The value of the square residual coefficient is between 0 and 1. The closer the square residual coefficient is to 1, the stronger the explanatory power of the independent variable on the output variable, and the better the degree of fit of the regression model. Conversely, the worse the degree of fit of the regression model is. The specific function is expressed as: R 2 =1-(SSR / SST) Among them, R 2 represents the square residual coefficient; SSR is the regression sum of squares, which represents the sum of squares of the difference between the model predicted value and the actual value; SST is the total sum of squares, which represents the sum of squares of the difference between the actual value and the mean; By calculating the global regression effects of characteristic indicators and accident losses through regression analysis, the ranking of the global regression effects among the multiple influencing factors of traffic accidents is obtained, and the indicator with the best regression effect is determined.

7. The method for predicting traffic accident losses based on machine learning and index screening according to claim 1, characterized in that: The S7 specifically includes the following contents: The weight of the number of indicators is introduced to calculate the importance of characteristic indicators to different accident indicators. The specific function is expressed as: Among them, importance(F) represents the global importance of the indicator; importance(f) represents the importance of the characteristic indicator corresponding to the accident indicator; N i Indicates the number of metrics in the metrics collection.

Images