Enterprise operation risk early warning method based on Benford factor and random forest model

Abstract

The invention belongs to the technical field of information security, and particularly relates to an enterprise operation risk early warning method based on a Benford factor and a random forest model. The method is characterized in that the random forest is used as a basic model, the advantages of the Benford law are fused, a data quality problem is converted into a classification problem, and finally the double effects of enterprise operation risk identification and early warning are achieved. Different from traditional enterprise risk prediction, a Benford factor is constructed to inspect data quality, then the Benford factor is fused into a random forest early warning model, the prediction precision of an operation risk early warning model is remarkably enhanced through the operation, risk early warning is carried out with the operation target of an agricultural product processing enterprise as an entry point, and the risk early warning efficiency is improved. In-depth analysis is carried out from the perspective, a set of operation risk early warning index system suitable for the industry is established, and the index system has better industry pertinence through financial and operation data verification of 36 months from 2020 to 2022.

Description

Technical Field

[0001] The present invention belongs to the field of information security technology, and in particular relates to an enterprise operation risk early warning method based on Benford factor and random forest model. Background Art

[0002] The quality of financial data is directly related to the accuracy of its prediction results. Frequent financial fraud incidents in listed companies have had a serious adverse impact on the market competition environment, and have also caused a significant decline in the credibility of accounting information. In the existing research on early warning of business risks, the close connection between the financial evaluation index system and corporate development is insufficient, and key financial indicators have not been fully and effectively reflected. Most of the current research has not fully considered the quality of financial indicators, which has brought certain limitations and uncertainties to the business risk early warning model we have constructed.

[0003] The efficacy coefficient method, also known as the efficacy function method, is a widely used quantitative evaluation method, which is widely used in business risk warning. The efficacy coefficient method can comprehensively evaluate the business risk of an enterprise and has high scientificity and accuracy. However, the traditional efficacy coefficient method also has its shortcomings, in order to avoid the drawbacks of the traditional efficacy analysis method. The entropy method is an objective weighting method, that is, by evaluating the information that can be given by each observation data, the weight of each indicator is determined. In information theory, entropy is a measure of uncertainty information.

[0004] Business analysis is a post-evaluation of the performance of an enterprise in the past accounting period based on the company's historical data. The early warning of business risks is to identify and control risks in a timely manner before they occur. Business risk early warning is a product of multidisciplinary integration. It is based on accounting, statistics and other disciplines, and screens out financial indicators based on the company's financial statements. The screened indicators are applied to the business risk early warning model of the enterprise to monitor the business situation of the enterprise in real time. After obtaining the early warning conclusion, the enterprise should accurately understand the current business situation faced by the enterprise, comprehensively analyze the key links, and adjust the operation strategy in a timely manner to ensure that business goals can be completed on time.

[0005] The Chinese invention patent with application number 201910209105.0 discloses a data fraud investigation method based on Benford's law, which includes: providing a data source, importing the environmental data of the data source and building an environmental data warehouse, using Benford's law to test the environmental data of the environmental data warehouse and outputting the test results, analyzing the test results, and being able to identify the regions or statistical links with greater suspicion of data fraud, identify the regions, enterprises and statistical indicators that need to focus on verifying the data quality, and provide technical support for improving the reliability and uncertainty analysis of data mining analysis results in related fields. "Quantitative Economics and Technical Economic Research" discloses a random forest model based on Benford's law and formulates an early warning of corporate operating risks. Research objective: The quality of financial data is crucial to the company's financial risk early warning. Financial data with high data quality often calculates a more stable financial risk early warning model. Benford's law can effectively evaluate the quality and characteristics of financial data, and construct a random forest model with Benford factors to deal with the impact of the quality of financial data on the financial risk early warning model, and can effectively improve the prediction accuracy of the financial early warning model. Research method: The quality of financial data is tested by Benford's law, Benford factors are constructed and added to financial indicator variables, and a random forest model based on Benford's law is established. The financial indicator data of listed companies in China's A-shares and the US stock market are selected for comparative empirical analysis, and the learning curve is used to tune the model parameters to determine the final model. The prediction effects of the random forest model based on Benford's law and the random forest model are compared. The study found that the Benford factor can identify specific sample points where financial fraud exists and provide information about data quality. Compared with the random forest model, the random forest model based on Benford's law can improve the accuracy of financial risk warning. Research innovation: Introduce Benford's law into the random forest model, construct Benford factors, and propose a random forest model based on Benford. Research value: The random forest model based on Benford's law can predict more accurately, increase the practicality of the random forest model, and provide a new perspective for the research on financial risk warning of listed companies. Summary of the invention

[0006] The purpose of the present invention is to study an enterprise business risk early warning method based on Benford factor and random forest model. On the basis of the existing technology, more accurate early warning is carried out, Benford factor is constructed and data quality is tested, and then it is integrated into the random forest early warning model. Based on the business objectives of agricultural product processing enterprises, a business risk early warning evaluation system suitable for the industry in this field is constructed to improve the accuracy of the enterprise business risk early warning model. At the same time, it helps researchers and practitioners to better understand the current business status of the enterprise, modify management decisions in time according to the business objectives, respond to business risks in time, and make the indicator system more industry-specific.

[0007] To achieve the above object, the present invention is implemented through the following technical solutions:

[0008] The enterprise operation risk early warning method based on Benford factor and random forest model is characterized by taking random forest as the basic model and integrating the advantages of Benford law, converting data quality problems into classification problems, and finally achieving the dual effects of operation risk identification and early warning. The specific steps of building the model are as follows:

[0009] 1) Check the data quality and construct the Benford factor Bj. Let the data set D(X1, Y1), (X2, Y2), ..., (Xn, Yn), where Xi represents the independent variable, Xi(Xi, 1, Xi, 2, ..., Xi, k)(i1, 2, 3, ...n), Yi represents the categorical variable, n represents the amount of data, and k represents the number of variables;

[0010] Then calculate the difference between the observed frequency and the theoretical frequency of the first digit and mark the Benford factor Bj;

[0011] 2) Add the constructed Benford factor as a new independent variable to the random forest model, and record the new variable as X iB (X i , 1, X i , 2, ..., X i , k, B i , 1, B i , 2, ..., B i , k)(i1, 2, 3, ...n), the data set is denoted as D B (X1 B , Y1), (X2 B , Y2),...,(X Bn , Y n );

[0012] 3) For dataset D B Resample and build a decision tree model, extract n sample data sets, denoted as DB(s) {s1, 2, ..., n), first set the initial number of decision trees according to the number of samples, and then optimize the model parameters;

[0013] 4) Finally, n decision trees are built for the n sample data sets selected, and the decision tree model sequence is recorded as {h1(X B(1) ), h2(X B(2) ), ..., hn(X B(n) )};

[0014] 5) Combine n decision trees to get a random forest model, hs(X B(s) ) represents the sth decision tree model, H(X B ) represents the combined random forest model, and the enterprise operation risk warning result indicators are obtained:

[0015]

[0016] Where I is the indicator function.

[0017] The enterprise operation risk warning result indicators are analyzed by Pearson correlation analysis using SPSS software. Representative indicators are extracted through analysis and screening, including but not limited to:

[0018] (1) Correlation analysis of debt-paying capacity indicators;

[0019] (2) Correlation analysis of profitability indicators;

[0020] (3) Correlation analysis of operating capacity indicators;

[0021] (4) Correlation analysis of development capability indicators;

[0022] (5) Correlation analysis of cigarette cost indicators.

[0023] The enterprise operation risk early warning result index uses the entropy method to calculate the weight, and the calculation method is:

[0024] Suppose there are n evaluation objects, each of which has m indicators, Xij is the jth indicator in the i-th year (i1, 2..., n; j1, 2..., m). The calculation steps of the entropy method can be divided into the following steps:

[0025] (1) First, construct the original data matrix. The formula is as follows:

[0026]

[0027] (2) The indicators are standardized as follows:

[0028] For the reverse indicator, the processing method is as follows:

[0029]

[0030] For positive indicators, the processing method is based on the formula

[0031]

[0032] For the moderate index, first use the formula X ij =-|X ij -a| (where a is the moderation value of the moderation index) converts it into a positive index and then performs dimensionless processing.

[0033] Afterwards, in order to avoid meaninglessness in calculating entropy and weight, negative numbers and zeros need to be non-negative.

[0034] Right now:

[0035] When Y ij ≤0, the coordinates need to be translated so that Y ij' >0, the processing method is based on the formula

[0036] Yij'=Yij+d

[0037] Where d is slightly larger than (Y ij )min integer.

[0038] Then we get the normalized matrix X Y :

[0039]

[0040] (3) Normalization processing, as shown below:

[0041]

[0042] (4) Calculate the entropy value, where k = ln1(n) > 0, as follows:

[0043]

[0044] (5) Calculate the differential sparseness, where e j >0, as shown in the following formula:

[0045] gj=1-ej

[0046] (6) The indicator weight is obtained as shown in the following formula

[0047]

[0048] W j It is the weight of the j-th indicator.

[0049] The business risk levels are divided into no warning, light warning, medium warning, heavy warning and huge warning, and the corresponding risk levels are no risk, light risk, medium risk, large risk and major risk.

[0050] The business risk level classification is completed based on the enterprise business risk warning result indicators, reflecting the scale and status of enterprise or industry risks.

[0051] The ANOVA method is used to verify the validity of the classification results of the business risk level classification.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1) Different from traditional enterprise risk prediction, the present invention constructs the Benford factor to test the data quality, and then integrates it into the random forest early warning model. This operation significantly enhances the prediction accuracy of the business risk early warning model. The present invention uses the business objectives of agricultural product processing enterprises as the starting point for risk warning, conducts in-depth analysis from this perspective, and establishes a set of business risk early warning indicator system suitable for the industry in this field, covering 36 months of financial and operational data from 2020 to 2022, making the indicator system more industry-specific.

[0054] 2) The present invention selects random forest as the basic model, and adopts entropy method, improved efficacy coefficient method and other methods to construct the business risk warning model of agricultural product processing enterprises, and combines Benford's law to conduct data quality inspection, forming an in-depth analysis and risk assessment of data in specific industries. Taking into account the timeliness of classification and better explanation of the reasons why enterprises fail to achieve their business goals, the improved efficacy coefficient method and random forest early warning model are adopted to successfully integrate the assessment and early warning of business risks, and prove the effectiveness of this comprehensive method in business risk research, and divide the risks into five levels for corresponding risk prediction. The improved efficacy coefficient method can comprehensively evaluate the business risks of enterprises, with high scientificity and accuracy, so as to avoid the drawbacks of traditional efficacy analysis methods.

[0055] 3) The business risk early warning system proposed in this invention is a comprehensive system that points out the importance of financial data quality to enterprise risk assessment and emphasizes the important role of industry characteristics in risk early warning. In addition, the model takes into account the quality factors of financial data, thereby improving the predictive ability of the business risk early warning model, converting data quality issues into classification issues, and ultimately achieving the dual effects of identifying and warning business risks of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The technical solution of the present invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0058] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the specific embodiments required to be used in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the specific embodiments described below are some implementation methods of the present invention. For ordinary technicians in this field, other specific embodiments can be obtained based on these specific embodiments without paying creative work.

[0059] The components of the embodiments of the present invention generally described and shown in the specific embodiments herein can be arranged and designed in countless different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the specific embodiments is not intended to limit the scope of the invention claimed, but merely represents selected embodiments of the present invention.

[0060] See Figure 1 , is a flow chart of an embodiment of the enterprise business risk early warning method based on Benford factor and random forest model of the present invention. With random forest as the basic model, it also integrates the advantages of Benford law, converts data quality problems into classification problems, and finally achieves the dual effects of business risk identification and early warning. The specific steps of building the model are as follows:

[0061] 1) Check the data quality and construct the Benford factor Bj. Let the data set be D(X1, Y1), (X2, Y2), ..., (Xn, Yn), where Xi represents the independent variable, Xi(Xi, 1, Xi, 2, ..., Xi, k)(i1, 2, 3, ...n), Yi represents the categorical variable, n represents the data volume, k represents the number of variables, and then calculate the difference between the observed frequency and the theoretical frequency of the first digit.

[0062] From the last column of Table 1 below, we can see that 12 financial indicators have undergone the χ2 goodness of fit test, and the χ2 test values ​​of 3 financial indicators exceed the critical value, indicating that their χ2 goodness of fit test failed, namely, quick ratio X2, accounts receivable turnover rate X5 and total asset turnover rate X6. This shows that there is a significant difference between the observed frequency of the first digit and the theoretical frequency of Benford's law, which points to potential financial fraud.

[0063] Table 1 Comparison of the observed and theoretical frequencies of the first digits of each indicator of a rice processing company

[0064]

[0065] 2) Add the constructed Benford factor as a new independent variable to the random forest model, and record the new variable as X iB (X i , 1, X i , 2, ..., X i , k, B i , 1, B i , 2, ..., B i , k)(i1, 2, 3, ...n), the data set is denoted as D B (X1 B , Y1), (X2 B , Y2),...,(X Bn , Y n );

[0066] 3) For dataset D B Resample and build a decision tree model, extract n sample data sets, denoted as D B(s) {s1, 2, ..., n), first set the initial number of decision trees according to the number of samples, and then optimize the model parameters.

[0067] 4) Finally, n decision trees are built for the n sample data sets selected, and the decision tree model sequence is recorded as {h1(X B(1) ), h2(X B(2) ), ..., hn(X B(n) )};

[0068] 5) Combine n decision trees to get a random forest model, hs(X B(s) ) represents the sth decision tree model, H(X B ) represents the combined random forest model, which indicates the early warning results of enterprise operation risks:

[0069]

[0070] Where I is the indicator function.

[0071] The RFB model contains 30 variables, including 15 financial indicator variables and 15 Benford factor variables. ( A random forest model (RFB) based on Benford's law was established. According to the principle of cross-validation, we split the selected samples into two parts, of which the training set accounted for 80%. In view of the sample size of the data set, we first set the number of decision trees to 100. The RFB model constructed using the financial data of an agricultural commercial enterprise had an initial prediction accuracy of 86.54%. The parameters of the model were adjusted in combination with the learning curve. Figure 1, the horizontal axis shows the number of decision trees, while the vertical axis shows the prediction accuracy of the model. We determine the best parameter options based on the parameter values ​​with higher prediction accuracy. In the range of 35 to 55, the prediction of the test set has a high accuracy. On this basis, the learning curve is further refined, and the results show that when the parameter value is 40, the accuracy of the test set reaches 89.21%. By adjusting the model parameters, the prediction accuracy is improved by 6.55%.

[0072] SPSS software was used to perform Pearson correlation analysis on the above-mentioned enterprise operation risk warning results. Through analysis and screening, representative indicators were extracted, including but not limited to:

[0073] (1) Correlation analysis of debt-paying capacity indicators;

[0074] The correlation analysis of the alternative indicators reflecting the debt repayment ability of a rice processing enterprise in 2020 is shown in Table 2. Note: ** is significant at the 0.01 level (two-tailed), and * is significant at the 0.05 level (two-tailed).

[0075] As can be seen from Table 2 below, the correlation coefficient between the cash flow to debt ratio and the quick ratio is greater than 0.8, which means that the two are highly correlated and can replace each other. Combined with the aforementioned indicator weights, the quick ratio indicator is retained. The correlation coefficient between the debt-to-asset ratio and the equity multiplier is greater than 0.8, so the equity multiplier with a higher weight is retained.

[0076] Table 22020 Debt Solvency Correlation Analysis

[0077]

[0078] (2) Correlation analysis of profitability indicators;

[0079] The correlation coefficient between the two optional indicators of the profitability index of a rice processing enterprise is calculated. As shown in Table 3 below, the correlation coefficient between the return on total assets and the return on net assets in the profitability index in 2020 is greater than 0.8. According to the weight value, the return on total assets is eliminated.

[0080] Table 32020 Profitability Correlation Analysis

[0081]

[0082] (3) Correlation analysis of operating capacity indicators;

[0083] By calculating the correlation coefficient between the two optional indicators of the operating capacity index of a rice processing enterprise, we can observe that in the operating capacity index in 2020, the correlation coefficient between the current asset turnover rate and the total asset turnover rate is greater than 0.8. For the sake of comprehensive weight comparison, we decided to retain the total asset turnover rate and eliminate the current asset turnover rate indicator, resulting in the following Table 4.

[0084] Table 42020 Operational Capacity Correlation Analysis

[0085]

[0086] (4) Correlation analysis of development capability indicators;

[0087] Calculating the correlation coefficient between the two optional indicators of a rice processing enterprise's development capability index, according to the data in Table 5 below, it can be observed that the correlation coefficient between the sustainable growth rate and the total asset growth rate is greater than 0.8. Based on this correlation finding and the previous weight comparison, we consider excluding the sustainable growth rate from the analysis and retaining the total asset growth rate.

[0088] Table 52020 Development Capacity Correlation Analysis

[0089]

[0090] (5) Correlation analysis of rice cost indicators.

[0091] The correlation coefficient between the two optional indicators of rice cost index of a rice processing enterprise is calculated. It can be seen from Table 6 below that the correlation coefficient between the two rice cost rates and the personnel cost rate is greater than 0.8. According to the weight comparison of the two, the two rice cost rates are greater than the weight of the personnel cost rate, so the personnel cost rate is eliminated. The correlation coefficient between per capita distribution efficiency and labor cost profit rate is greater than 0.8. According to the weight comparison of the two, the labor cost profit rate is retained.

[0092] Table 6 Correlation analysis of rice cost indicators in 2020

[0093]

[0094] The entropy method is used to weight the selected indicators. The weights of the enterprise operation risk warning result indicators are calculated using the entropy method. The calculation process is as follows:

[0095] Suppose there are n evaluation objects, each of which has m indicators, Xij is the jth indicator in the i-th year (i1, 2..., n; j1, 2..., m). The calculation steps of the entropy method can be divided into the following steps:

[0096] (1) First, construct the original data matrix. Formula (2) is as follows:

[0097]

[0098] (2) The indicators are standardized as follows:

[0099] For the reverse indicator, the processing method is as follows:

[0100]

[0101] For positive indicators, the processing method is based on formula (4)

[0102]

[0103] For the moderate index, first use the formula X ij =-|X ij -a| (where a is the moderation value of the moderation index) converts it into a positive index and then performs dimensionless processing.

[0104] Afterwards, in order to avoid meaninglessness in calculating entropy and weight, negative numbers and zeros need to be non-negative.

[0105] Right now:

[0106] When Y ij ≤0, the coordinates need to be translated so that Y ij' >0, the processing method is based on formula (5)

[0107] Yij'=Yij+d Formula (5)

[0108] Where d is slightly larger than (Y ij )min integer.

[0109] Then we get the normalized matrix X Y :

[0110]

[0111] (3) Normalization processing, as shown in the following formula (7):

[0112]

[0113] (4) Calculate the entropy value, where k = ln1(n) > 0, as shown in the following formula (8):

[0114]

[0115] (5) Calculate the differential sparseness, where ej>0, as shown in the following formula (9):

[0116] gj=1-ej Formula (9)

[0117] (6) The obtained indicator weight is the weight of the jth indicator, as shown in the following formula (10):

[0118]

[0119] W j It is the weight of the j-th indicator.

[0120] After the above calculations, the specific calculation results of the entropy value and weight of the financial early warning indicators from 2020 to 2022 are presented in Table 7 below.

[0121] Table 7 Entropy and weight of financial early warning indicators from 2020 to 2022

[0122]

[0123] According to the results of the early warning of the business risk of the enterprise, the business risk level is divided into no warning, light warning, medium warning, heavy warning, and great warning. The corresponding risk levels are no risk, light risk, medium risk, large risk, and major risk, which are represented by green light, blue light, yellow light, orange light, and red light respectively. This study evaluates the business risk of the enterprise according to the completion of the business objectives of the enterprise, and then divides the risk level and establishes a business risk assessment level division table, as shown in Table 8 below.

[0124] Table 8 Classification of business risk assessment levels

[0125]

[0126] Through the calculation principle of the efficacy coefficient method, a comprehensive score is calculated for the enterprise's monthly risk, and then the enterprise's risk level for that month is determined based on the score range. The results of the monthly risk levels from 2020 to 2022 are shown in Table 9 below.

[0127] Table 9 Risk warning results of a rice processing enterprise from 2020 to 2022

[0128]

[0129] The effectiveness of the business risk classification is verified by using the ANOVA method to verify the effectiveness of the classification results. The final verification results of this embodiment are:

[0130] Combined with the correlation analysis of rice cost indicators from 2020 to 2022, the personnel cost rate indicator and per capita distribution efficiency were finally eliminated, and the growth capacity indicators entering the efficiency coefficient method were the two rice cost rates, labor cost profit rate, single box logistics cost growth rate, inter-provincial transaction ratio, rice sales growth rate and key brand single box sales growth rate. After correlation testing and indicator analysis, the financial early warning indicator system of the rice processing enterprise was finally determined, as shown in Table 10 below.

[0131] Table 10 Financial early warning indicator system of a rice processing enterprise

[0132]

[0133] Establishing an early warning system for business risks and integrating big data technology can help managers and related personnel to discover potential business risks in a timely manner. It can help corporate leaders integrate internal resources more effectively and assist companies in making wise strategic decisions, thereby driving the company forward in the right direction. This step significantly enhances the company's risk management capabilities so that the company can better adapt to evolving market conditions.

[0134] Although embodiments of the present invention have been shown and described above, it will be understood by those skilled in the art that various changes, modifications, substitutions and extensions may be made to the embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The enterprise operation risk early warning method based on Benford factor and random forest model is characterized by: Using random forest as the basic model and integrating the advantages of Benford's law, the data quality problem is converted into a classification problem, and finally the dual effects of business risk identification and early warning are achieved. The specific steps of building the model are as follows: 1) Check the data quality and construct the Benford factor Bj. Let the data set D(X1, Y1), (X2, Y2), ..., (Xn, Yn), where Xi represents the independent variable, Xi(Xi, 1, Xi, 2, ..., Xi, k)(i 1, 2, 3, ... n), Yi represents the categorical variable, n represents the amount of data, and k represents the number of variables; Then calculate the difference between the observed frequency and the theoretical frequency of the first digit and mark the Benford factor Bj; 2) Add the constructed Benford factor as a new independent variable to the random forest model, and record the new variable as X iB (X i , 1, X i , 2, ..., X i , k, B i , 1, B i , 2, ..., B i , k)(i 1, 2, 3, ... n), the data set is denoted as D B (X1 B , Y1), (X2 B , Y2),...,(X Bn , Y n ); 3) For dataset D B Resample and build a decision tree model, extract n sample data sets, denoted as D B(s) {s 1, 2, ..., n), first set the initial number of decision trees according to the number of samples, and then optimize the model parameters; 4) Finally, n decision trees are built for the n sample data sets selected, and the decision tree model sequence is recorded as {h1(X B (1) ), h2(X B(2) ), ..., hn(X B(n) )}; 5) Combine n decision trees to get a random forest model, hs(X B(s) ) represents the sth decision tree model, H(X B ) represents the combined random forest model, and the enterprise operation risk warning result indicators are obtained: Where I is the indicator function.

2. The enterprise operation risk early warning method based on Benford factor and random forest model according to claim 1 is characterized in that: The enterprise operation risk warning result indicators are analyzed by Pearson correlation analysis using SPSS software. Representative indicators are extracted through analysis and screening, including but not limited to: (1) Correlation analysis of debt-paying capacity indicators; (2) Correlation analysis of profitability indicators; (3) Correlation analysis of operating capacity indicators; (4) Correlation analysis of development capability indicators; (5) Correlation analysis of cigarette cost indicators.

3. The enterprise operation risk early warning method based on Benford factor and random forest model according to claim 1 is characterized in that: The enterprise operation risk early warning result index uses the entropy method to calculate the weight, and the calculation method is: Suppose there are n evaluation objects, each of which has m indicators, Xij is the jth indicator in the i-th year (i1, 2..., n; j1, 2..., m), and the calculation steps of the entropy method can be divided into the following steps: (1) First, construct the original data matrix. The formula is as follows: (2) The indicators are standardized as follows: For the reverse indicator, the processing method is as follows: For positive indicators, the processing method is based on the formula For the moderate index, first use the formula X ij =-|X ij -a| (where a is the appropriate value of the appropriateness index) converts it into a positive index and then performs dimensionless processing; Afterwards, in order to avoid meaninglessness in calculating entropy and weight, negative numbers and zeros need to be non-negative, that is: When Y ij ≤0, the coordinates need to be translated so that Y ij' >0, the processing method is based on the formula Yij'=Yij+d Where d is slightly larger than (Y ij )min integer; Then we get the normalized matrix X Y : (3) Normalization processing, as shown below: (4) Calculate the entropy value, where k = ln1(n) > 0, as follows: (5) Calculate the differential sparseness, where e j >0, as shown in the following formula: gj=1-ej (6) The indicator weight is obtained as shown in the following formula W j It is the weight of the j-th indicator.

4. The enterprise operation risk early warning method based on Benford factor and random forest model according to claim 1 is characterized in that: The business risk levels are divided into five levels: no warning, light warning, medium warning, heavy warning and huge warning, and the corresponding risk levels are no risk, light risk, medium risk, large risk and major risk.

5. The enterprise operation risk early warning method based on Benford factor and random forest model according to claim 4 is characterized in that: The business risk level classification is completed based on the enterprise business risk warning result indicators, reflecting the scale and status of enterprise or industry risks.

6. The enterprise operation risk early warning method based on Benford factor and random forest model according to claim 5 is characterized by: The ANOVA method is used to verify the validity of the classification results of the business risk level classification.

Images