Financial risk identification method based on machine learning

Abstract

According to the financial risk identification method based on machine learning, financial statement whitewash is identified and deeply analyzed, financial counterfeiting feature label data of supervision notification is mined, enterprise whitewash risks and operation risks are researched and judged by means of artificial intelligence technologies of a decision tree and an early warning system, customer financial risk portraits are enriched, and the financial risk identification efficiency is improved. A financial analysis report is automatically generated, a multi-dimensional risk analysis basis is provided for post-loan management of a customer manager, a personnel manager and an operation main responsible person, and financial risk intelligent analysis and multi-dimensional risk information service capabilities are improved.

Description

Technical Field

[0001] The present invention belongs to the field of financial processing, and in particular relates to a financial risk identification method based on machine learning. Background Art

[0002] At present, my country's corporate financial fraud detection and financial and operational risk analysis are mainly based on manual experience and require professional personnel to conduct analysis. However, the grassroots account managers in general banks are young, unstable, and under great marketing pressure. They lack sufficient time and experience to identify financial fraud risks and cannot achieve true intelligence and automation.

[0003] At present, the existing machine learning models with a small number of features were born, which can realize preliminary detection of financial data. Commonly used methods include: step S1, obtaining the financial data of listed companies in various industries for training; step S2, using the SMOTE method to oversample the acquired data; step S3, performing feature deletion and using the principal component analysis method to reduce the feature dimension; step S4, building four basic models: logistic regression model, support vector machine model, decision tree model, neural network model; and integrated model: Stacking algorithm model; step S5, parameter selection through grid search and K-fold cross validation; step S6, inputting the processed training data into the machine learning model, training the machine learning model, and obtaining a trained machine learning model; step S7, obtaining the financial data to be identified, inputting the financial data to be identified into the machine learning model, and outputting the results of the fraudulent financial data of listed companies.

[0004] However, the above technical features have obvious shortcomings: 1. Data limitations: Since only annual report data is used for model training and prediction, the existing technology does not analyze whether there are financial fraud and operating risks in quarterly and semi-annual reports; 2. Insufficient feature engineering: Due to the small number of derivative variables used, the model cannot fully explore the hidden information and patterns in the financial data, which limits the model's predictive performance and generalization ability; 3. Model entry index screening rules: The current technology's screening rules for model entry indicators are not reasonable enough, which affects the effectiveness and stability of the model. This may cause the model to perform poorly when faced with complex or abnormal situations; 4. The definition of bad samples is vague: The number of bad samples is small and the definition is vague, which limits the generalization ability of the model. At the same time, this may also cause the model to deviate when dealing with actual situations; 5. The output results are relatively simple: only the model results are output. The output results are relatively simple, without strong persuasiveness and practicality.

[0005] Therefore, it is particularly important to design a new machine learning model to process financial data, truly intelligently and automatically identify financial fraud risks, and present the company's financial fraud risks and operating risks in the form of comprehensive report interpretation. Summary of the invention

[0006] In order to solve the above problems, this application designs a financial risk identification method based on machine learning, which automatically generates financial analysis reports by identifying and deeply analyzing financial statement fraud, thereby improving the financial risk intelligent analysis and multi-dimensional risk information service capabilities.

[0007] A financial risk identification method based on machine learning, comprising:

[0008] Step S1, collect the test data set required by the model;

[0009] Step S2, divide the test data set into four categories: classic financial indicators, empirical financial indicators, whitewashing indicators, and Themis indicators, and construct 280 derived financial indicators;

[0010] Step S3, analyzing the importance of the derived financial indicators, screening out the derived financial indicators with high feature importance ranking and information value IV exceeding 0.02, and ensuring that their trends are interpretable;

[0011] Step S4: input the screened derivative financial indicator data and the corresponding financial whitewashing labels into the XGBoost model for training, and adjust the training parameters of XGBoost to obtain the financial whitewashing model; input the screened derivative financial indicator data and the corresponding operating risk labels into the XGBoost model for training, and adjust the training parameters of XGBoost to obtain the operating risk model;

[0012] Step S5: construct a cross-checking relationship module by presetting the cross-checking relationship between multiple financial basic indicators; when the financial report basic indicators input by the enterprise do not meet the preset cross-checking relationship, the system will display and output corresponding descriptive text;

[0013] Step S6, establish an early warning indicator module and set a threshold; when the enterprise inputs financial data and industry information, the early warning indicator module calculates the enterprise early warning indicator value and compares it with the threshold. If the threshold is triggered, the early warning indicator module will output the triggered early warning indicator, specific data, risk description and matters of concern;

[0014] Step S7: Establish a balance sheet display module to obtain the company's debt situation, and use text description and chart display to analyze key financial indicators and analyze the asset composition structure in detail;

[0015] Step S8: Input the enterprise's financial data into the financial whitewashing model to analyze the enterprise's financial whitewashing from three perspectives: inflated revenue, reduced costs, and inflated investment returns;

[0016] Step S9: Input the enterprise's financial data into the business risk model, and analyze the enterprise's business risk from five perspectives: debt repayment ability, profitability, operating ability, growth ability, and cash flow.

[0017] Furthermore, the test data set includes: financial samples and business samples;

[0018] The financial samples include: good financial samples and bad financial samples;

[0019] The financial bad samples include: collecting administrative penalty data of listed companies in the past ten years from the website of the China Securities Regulatory Commission, and then parsing the penalty information texts, defining listed companies that have been punished for financial whitewashing and false financial reports as bad customers; for these bad customers, using API data sources to collect their reported current financial reports as bad samples, and paying special attention to the fact that these data are not revised financial reports;

[0020] The good financial samples include: defining listed companies that have not been punished by the China Securities Regulatory Commission for financial fraud as good customers; for these good customers, using API data sources to collect their financial reports for the past five years as good samples;

[0021] The business samples include: good business samples and bad business samples;

[0022] The good operating samples include: collecting key information such as the financial report data, enterprise size, industry, and five-level post-loan classification of the company's customers; screening out customers who have been in a state of concern or bad in the past four years as bad customers, and selecting the financial reports of these bad customers in the six months before they entered the state of concern or bad as bad samples;

[0023] The bad operating samples include: customers who have had credit in the bank for nearly four years and whose five-level classification is normal are defined as good customers, and the financial statements of these good customers are randomly selected as good samples.

[0024] Furthermore, in step S5, there are 9 cross-references, specifically including:

[0025] Monetary funds at the end of the period ≠ monetary funds at the beginning of the period + net increase in cash and cash equivalents at the end of the period;

[0026] Undistributed profit at the end of the period > Undistributed profit at the beginning of the period + Net profit at the end of the period;

[0027] Total assets ≠ total liabilities + total owner's equity;

[0028] Net cash flow from operating activities ≠ Subtotal of cash inflow from operating activities - Subtotal of cash outflow from operating activities;

[0029] Net cash flow from investing activities ≠ Subtotal of cash inflows from investing activities - Subtotal of cash outflows from investing activities;

[0030] Net cash flow from financing activities ≠ Subtotal of cash inflows from financing activities - Subtotal of cash outflows from financing activities;

[0031] Net cash increase ≠ net cash flow from operating activities + net cash flow from investing activities + net cash flow from financing activities;

[0032] Total comprehensive income ≠ net profit + other comprehensive income;

[0033] Net profit ≠ total profit - income tax expense.

[0034] Furthermore, the step S6 specifically includes:

[0035] Step S61: Based on the sample data of each industry, draw an IV chart of indicators applicable to nine industries, including electricity, heat, gas and water production and supply, real estate, construction, transportation, warehousing and postal services, wholesale and retail, water conservancy, environment and public facilities management, information transmission, software and information technology services, manufacturing, leasing and business services, as well as all industries; divide the derived financial indicators into boxes, and draw a corresponding IV chart with each group as the horizontal axis and the number and bad sample rate of each group as the double vertical axes;

[0036] Step S62: Combine the IV graph trend to select early warning indicators with early warning significance;

[0037] Step S63: Determine the optimal threshold of the warning indicator based on the selected warning indicators by combining the precision rate and the alarm trigger rate;

[0038] Precision rate = the number of bad customers triggered by a single warning indicator / the number of customers triggered by a single warning indicator. The precision rate evaluates whether the warning indicator can effectively warn, which requires more than or equal to twice the number of bad samples in the industry;

[0039] Trigger rate = number of customers triggered by a single warning indicator / total number of customers. The trigger rate evaluates the alarm ratio of the warning indicator and should not be too large, less than or equal to 0.1;

[0040] Step S64: Determine the warning level according to the precision of the customer whose indicator triggers the alarm;

[0041] Step S65: Output the risk description and concerns of the warning indicators.

[0042] Furthermore, in step S8, the analysis of inflated income includes: analyzing from two dimensions, the trend of changes in accounts receivable and main business income, the trend of changes in non-operating income and total operating income, and comparing the difference between the current value and the same period last year; if the difference fluctuates too much and exceeds the preset standard threshold, the financial whitewashing model will issue a risk warning;

[0043] The trend of changes in accounts receivable and main business income mainly analyzes the Δ accounts receivable / Δ main business income indicator;

[0044]

[0045] The Δ accounts receivable / Δ main business income indicator is used to evaluate the reasonableness of the year-on-year change in accounts receivable and the year-on-year change in main business income. The reasonable threshold of this indicator is [-∞,1.5]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating the main business income by inflating accounts receivable. When Δ accounts receivable / Δ main business income>=1.5, it is prompted: "Beyond the normal fluctuation range, there is unreasonable situation in the change of accounts receivable or main business income. It is recommended to pay attention to the quality and rationality of accounts receivable and main business income". In other cases, it is prompted: "Within the normal fluctuation range";

[0046] The trend of changes in non-operating income and total operating income is analyzed by analyzing the growth rate of non-operating income and the proportion of non-operating income to total operating income. If the growth rate of non-operating income is greater than 1 and the proportion of non-operating income to total operating income is greater than or equal to 0.05, it is prompted that "non-operating income grows rapidly and accounts for a high proportion of total operating income. There are unreasonable situations in the changes of non-operating income and total operating income. It is recommended to pay attention to the quality and rationality of non-operating income and total operating income." In other cases, it is prompted that "non-operating income and total operating income show a normal trend of change";

[0047] When any of the conclusions in the above two parts is unreasonable, the conclusion is "the risk of inflated income is high", otherwise "the risk of inflated income is low".

[0048] Furthermore, in step S8, the analysis of the virtual reduction of costs includes: analyzing the key indicators of the trend of changes in accounts payable & prepaid and main business income, and the trend of changes in main business costs and main business income, and performing text interpretation and chart display;

[0049] The main analysis of the trend of changes in accounts payable & prepaid and main business income is △ accounts payable & prepaid / △ main business income indicators,

[0050]

[0051] The trend indicator of accounts payable & prepaid and main business income is intended to evaluate the reasonableness of the year-on-year change in accounts payable & prepaid and the year-on-year change in main business income. The reasonable threshold of this indicator is [-2.5,∞]; the greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of falsely reducing costs by falsely reducing accounts payable & prepaid; when △ accounts payable & prepaid / △ main business income <= -2.5, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in accounts payable & prepaid or main business income. It is recommended to pay attention to the quality and rationality of accounts payable & prepaid and main business income." Other situations are prompted: within the normal fluctuation range;

[0052] The trend of changes in the main business cost and main business income mainly analyzes the difference between the main business cost / main business income in the current period and the same period last year. When the value is within [-0.3, 0.3], it is "in the normal fluctuation range". When the value is <-0.3, it indicates that "the downward trend is more significant compared with the previous year, which requires special attention". When the value is >0.3, it indicates that "the upward trend is more significant compared with the previous year, which requires special attention". When the value is == 0, it indicates that "it remains unchanged compared with the previous year";

[0053] When any of the conclusions in the above two parts is unreasonable, the conclusion “the risk of false cost reduction is high” is given; otherwise, “the risk of false cost reduction is low” is given.

[0054] Furthermore, in step S8, the analysis of inflated investment returns includes:

[0055] Analyze the changing trends of long-term equity investment and main business income, and the changing trends of investment income and operating profit;

[0056] Calculation formula for the trend of long-term equity investment and main business income:

[0057]

[0058] The long-term equity investment and main business income change trend indicator is designed to evaluate the year-on-year change in long-term equity investment and the reasonableness of the year-on-year change in main business income. The reasonable threshold of this indicator is [-∞, 0.8]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating investment returns through fictitious growth period equity investment. When △ long-term equity investment / △ main business income > = 0.8, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in long-term equity investment or main business income. It is recommended to pay attention to the quality and rationality of long-term equity investment and main business income." Other situations are prompted: within the normal fluctuation range;

[0059] Calculation formula for investment income and operating profit trend:

[0060]

[0061] The △Investment income / △Operating profit indicator is designed to assess the reasonableness of the year-on-year change in investment income and the year-on-year change in operating profit. The reasonable threshold of this indicator is [-∞, 0.5]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating investment income. When △Investment income / △Operating profit>=0.5, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in investment income or operating profit. It is recommended to pay attention to the quality and rationality of investment income and operating profit." In other cases, it is prompted: It is within the normal fluctuation range;

[0062] When any of the conclusions in the above two parts is unreasonable, the conclusion is "the risk of inflated investment returns is high", otherwise "the risk of inflated investment returns is low".

[0063] The advantages and effects of this application are as follows:

[0064] This application designs a financial risk identification method based on machine learning. Through the identification and in-depth analysis of financial statement fraud, it mines the characteristic label data of financial fraud reported by the regulator, and uses the artificial intelligence technology of decision trees and early warning systems to judge the company's fraud risk and operating risk, enrich the customer's financial risk portrait, and automatically generate financial analysis reports. It provides a multi-dimensional risk analysis basis for account managers, personnel managers, and business owners to conduct post-loan management, and enhances the ability of intelligent analysis of financial risks and multi-dimensional risk information services.

[0065] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application so that it can be implemented in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following is a detailed description of the preferred embodiments of the present application in conjunction with the accompanying drawings as follows.

[0066] Based on the detailed description of the specific embodiments of the present application in combination with the accompanying drawings below, those skilled in the art will become more aware of the above and other objects, advantages and features of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings without creative work. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale.

[0068] Figure 1A flowchart of a machine learning-based financial risk identification method designed for this application. DETAILED DESCRIPTION

[0069] To make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the present application. Obviously, the described embodiment is a part of the embodiment of the present application, rather than all of the embodiments. In the following description, specific details such as specific configuration and components are provided only to help fully understand the embodiments of the present application. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, for clarity and brevity, the description of known functions and structures is omitted in the embodiment.

[0070] It should be understood that the references to "one embodiment" or "this embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present application. Therefore, the references to "one embodiment" or "this embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0071] In addition, the present application may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplicity and clarity, and does not in itself indicate the relationship between the various embodiments and / or settings discussed.

[0072] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" in this article describes another type of association object relationship, indicating that there can be two relationships. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0073] The term "at least one" in this article is merely a description of the association relationship of associated objects, indicating that there may be three relationships. For example, at least one of A and B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0074] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusions.

[0075] Example 1

[0076] This example mainly introduces a specific design of a financial risk identification method based on machine learning. Figure 1 , including the following steps:

[0077] Step S1, collect the test data set required by the model;

[0078] The test data set required for training the financial whitewashing model is: financial samples.

[0079] The financial samples include: good financial samples and bad financial samples; they are mainly achieved through the following steps: First, collect the administrative penalty data of listed companies in the past ten years from the website of the China Securities Regulatory Commission, and then parse the penalty information texts, and define the listed companies that have been punished for financial whitewashing and false financial reports as bad customers. For these bad customers, use the API data source to collect their current financial reports that have been notified as bad financial samples, and pay special attention to the fact that these data are not revised financial reports. At the same time, the listed companies that have not been punished by the China Securities Regulatory Commission for financial whitewashing are defined as good customers. For these good customers, use the API data source to collect their financial reports in the past five years as good financial samples.

[0080] The test data set required for training the business risk model is: business samples.

[0081] The business samples include: good business samples and bad business samples; mainly achieved through the following steps: First, collect key information such as the financial report data, enterprise scale, industry, and post-loan five-level classification level of the company's customers. Then, screen out customers who have been in a state of concern or bad in the past four years as bad customers, and select the financial reports of these bad customers half a year before they entered the state of concern or bad as bad business samples. At the same time, customers who have been granted credit in the bank for nearly four years and whose five-level classification is normal are defined as good customers, and the financial reports of these good customers are randomly selected as good business samples. Through this method, the accuracy and reliability of the data can be ensured, providing strong support for the training of the business risk model.

[0082] Step S2, divide the test data set into four categories: classic financial indicators, empirical financial indicators, whitewashing indicators, and Themis indicators, and construct 280 derived financial indicators;

[0083] The derived financial indicators can be derived based on classic financial indicators in the industry, the experience of financial experts, and relevant financial books and papers. These derived indicators ensure the interpretability and practicality of the model and provide strong support for subsequent model construction.

[0084] Step S3, analyzing the importance of the derived financial indicators, screening out the derived financial indicators with high feature importance ranking and information value IV exceeding 0.02, and ensuring that their trends are interpretable;

[0085] Step S4: input the screened derivative financial indicator data and the corresponding financial whitewashing labels into the XGBoost model for training, and adjust the training parameters of XGBoost to obtain the financial whitewashing model; input the screened derivative financial indicator data and the corresponding operating risk labels into the XGBoost model for training, and adjust the training parameters of XGBoost to obtain the operating risk model;

[0086] The screened derivative financial indicator data are input into the XGBoost model for training, and the training parameters of XGBoost are adjusted to obtain a financial whitewashing model; and the risk level (AE) is set for the XGBoost model to obtain an operating risk model; wherein the derivative financial indicator data used in the financial whitewashing model are financial samples, and the derivative financial indicator data used in the operating risk model are operating samples.

[0087] When there is a new financial report that needs to be analyzed, just enter the model indicators into these two models to obtain the corresponding risk level as the first part of the report.

[0088] Step S5: construct a cross-checking relationship module by presetting the cross-checking relationship between multiple financial basic indicators; when the financial report basic indicators input by the enterprise do not meet the preset cross-checking relationship, the system will display and output corresponding descriptive text;

[0089] The cross-check relationship module has 11 preset cross-check relationships to ensure the accuracy and consistency of financial data. Through this module, you can quickly find abnormalities in the data and provide prompts and warnings. The preset cross-check relationships and corresponding descriptions are as follows:

[0090]

[0091]

[0092] Step S6: Establish an early warning indicator module and set a threshold. When the enterprise inputs financial data and industry information, the early warning indicator module calculates the enterprise early warning indicator value and compares it with the threshold. If the threshold is triggered, the early warning indicator module will output the triggered early warning indicator, specific data, risk description and matters of concern. Specifically, it includes:

[0093] Step S61, based on the sample data of each industry, draw the indicator IV graph applicable to nine industries, including electricity, heat, gas and water production and supply, real estate, construction, transportation, warehousing and postal services, wholesale and retail, water conservancy, environment and public facilities management, information transmission, software and information technology services, manufacturing, leasing and business services, as well as all industries; for industries with a small number of samples, merge them according to the similarity between industries, and perform equal frequency binning or chi-square binning on the derived financial indicators, and draw the corresponding IV graph with each group as the horizontal axis and the number of each group and the bad sample rate as the double vertical axes; the IV graph can intuitively represent the trend of the positive example proportion corresponding to the change of the feature value.

[0094] Step S62, combining the IV graph trend, screen out warning indicators with warning significance; for example, the lower the net profit margin, the higher the risk, the IV graph trend should be that as the net profit margin increases, the bad sample rate decreases, and the curve shows a trend from top to bottom.

[0095] Step S63: Determine the optimal threshold of the warning indicator based on the selected warning indicators by combining the precision rate and the alarm trigger rate;

[0096] Precision rate = the number of bad customers triggered by a single warning indicator / the number of customers triggered by a single warning indicator. The precision rate evaluates whether the warning indicator can effectively warn, which requires more than or equal to twice the number of bad samples in the industry;

[0097] Trigger rate = number of customers triggered by a single warning indicator / total number of customers. The trigger rate evaluates the alarm ratio of the warning indicator and should not be too large, less than or equal to 0.1;

[0098] Step S64: Determine the warning level according to the precision of the customer whose indicator triggers the alarm;

[0099] Step S65: Output the risk description and concerns of the warning indicators.

[0100] Through the above steps, this application successfully built an early warning indicator module. When there is a new financial report that needs to be analyzed, you only need to enter the financial data and industry information, and the system will automatically calculate the preset early warning indicator value and determine the set threshold. If the threshold is triggered, the system will output the triggered early warning indicator, specific data, risk description and concerns. This module helps users to identify potential risks in a timely manner and provide strong support for decision-making. The following are some examples. In actual scenarios, early warning indicators can be screened based on actual data.

[0101]

[0102]

[0103]

[0104]

[0105] Step S7, establish a balance sheet display module to obtain the company's liabilities; it can display the company's assets, liabilities and owner's equity in detail. This module is divided into two parts: text description and chart display. It not only provides a detailed analysis of key financial indicators, but also deeply analyzes the asset composition structure. The top three items of current assets and non-current assets will be described in text, items and proportions. Liabilities will analyze the composition structure of current liabilities and non-current liabilities, and the top three items will be described in text, items and proportions. More importantly, this module also provides the year-on-year changes of each item compared with the previous year to help users fully understand the financial status of the company.

[0106] In the text description, the subjects with large fluctuations compared with the previous year will be pointed out. Here, the subjects with fluctuations exceeding 30% will be pointed out to provide key information for users. In the table, special colors will be used to indicate these subjects with large fluctuations, making important information more prominent.

[0107] This module is characterized by its comprehensiveness and detail. Through the combination of text and charts, it provides users with a comprehensive and intuitive balance sheet view, which helps to better understand the financial status and asset structure of the company. The following is a sample template:

[0108] Asset analysis: As of the end of {December 2020}, the company's total assets were {638.2152} million yuan, of which current assets were {454.8842} million yuan, mainly {accounts payable}, {inventory}, and {other receivables}, accounting for {54.08%}, {21.33%}, and {9.75%} of current assets, respectively. Non-current assets were {183.3310} million yuan, mainly {long-term equity investments}, {fixed assets}, and {construction in progress}, accounting for {54.73%}, {23.49%}, and {21.73%} of non-current assets, respectively. Among them, {inventory}, {other receivables}, {notes receivable}, and {monetary funds} items fluctuated by more than 30% compared with the previous year. It is recommended to focus on the reasons for the large changes in the above items.

[0109] Liabilities and owners' equity analysis: As of the end of {December 2020}, the company's total liabilities were {444.8691 million}, of which current liabilities were {444.8691 million}, mainly {short-term loans}, {accounts payable}, and {other payables}, accounting for {36.64%}, {36.24%}, and {23.82%} of current liabilities, respectively. Non-current liabilities were {0.00} million. Owners' equity was {193.3461} million, and paid-in capital was {50 million. Among them, {short-term loans}, {notes payable}, and {other payables} items fluctuated by more than 30% compared with the previous year. It is recommended to focus on the reasons for the large changes in the above items.

[0110]

[0111] Step S8: Input the enterprise's financial data into the financial fraud model, analyze the enterprise's financial fraud from three perspectives: inflated revenue, reduced costs, and inflated investment returns; and provide a comprehensive interpretation of the enterprise's true financial situation.

[0112] It is mainly divided into three parts:

[0113] 1. Analysis of inflated income: mainly analyze from two dimensions, the trend of changes in accounts receivable and main business income, the trend of changes in non-operating income and total operating income, and compare the difference between the current value and the same period last year. If the difference fluctuates too much and exceeds the preset standard threshold, the system will issue a risk warning.

[0114] The trend of accounts receivable and main business income is mainly analyzed by the following indicators, and the calculation formula is:

[0115] The Δ accounts receivable / Δ main business income indicator is used to evaluate the reasonableness of the year-on-year change in accounts receivable and the year-on-year change in main business income. The reasonable threshold of this indicator is [-∞,1.5]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating the main business income by inflating accounts receivable. When Δ accounts receivable / Δ main business income>=1.5, it is prompted: "Beyond the normal fluctuation range, there is unreasonable situation in the change of accounts receivable or main business income. It is recommended to pay attention to the quality and rationality of accounts receivable and main business income". In other cases, it is prompted: "Within the normal fluctuation range";

[0116] The trend of changes in non-operating income and total operating income is analyzed by analyzing the growth rate of non-operating income and the proportion of non-operating income to total operating income. If the growth rate of non-operating income is greater than 1 and the proportion of non-operating income to total operating income is greater than or equal to 0.05, it is prompted that "non-operating income grows rapidly and accounts for a high proportion of total operating income. There are unreasonable situations in the changes of non-operating income and total operating income. It is recommended to pay attention to the quality and rationality of non-operating income and total operating income." In other cases, it is prompted that "non-operating income and total operating income show a normal trend of change";

[0117] When any of the conclusions in the above two parts is unreasonable, the conclusion is "the risk of inflated income is high", otherwise "the risk of inflated income is low".

[0118] The text interpretation template is as follows:

[0119] Combined with the analysis results of the financial whitewashing model, it is preliminarily determined that the risk of inflated revenue in the company's {2020 financial report} is {low}, mainly analyzed from the following two dimensions:

[0120] Trends in changes in accounts receivable and main business income: As of the end of {December 2020}, the ratio of the year-on-year change in accounts receivable to the year-on-year change in main business income of the company was {-0.84}, while the ratio in the same period last year was {0.87}. The difference between the two periods (△accounts receivable / △main business income) was {-1.72}, which is {within the normal fluctuation range}.

[0121] Trend of changes in non-operating income and total operating income: As of the end of {December 2020}, the company's non-operating income was {1.1550} million yuan, and non-operating income accounted for {0.16%} of total income. Non-operating income and total operating income showed a normal trend of change.

[0122]

[0123] 2. Analysis of false cost reduction: by analyzing the key indicators of the trend of changes in accounts payable & prepaid and main business income, and the trend of changes in main business costs and main business income, and providing textual interpretation and chart presentation;

[0124] This section will dig deep into whether the company has engaged in cost-reduction activities. By analyzing the key indicators of the changing trends of accounts payable and prepaid expenses and main business income, and the changing trends of main business costs and main business income, text interpretation and chart presentation will be provided to help users identify potential risk points.

[0125] Calculation formula for the trend of accounts payable & prepaid expenses and main business income

[0126]

[0127] The trend indicator of accounts payable & prepaid and main business income is intended to evaluate the reasonableness of the year-on-year change in accounts payable & prepaid and the year-on-year change in main business income. The reasonable threshold of this indicator is [-2.5,∞]; the greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of falsely reducing costs by falsely reducing accounts payable & prepaid; when △ accounts payable & prepaid / △ main business income <= -2.5, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in accounts payable & prepaid or main business income. It is recommended to pay attention to the quality and rationality of accounts payable & prepaid and main business income." Other situations are prompted: within the normal fluctuation range;

[0128] The trend of changes in the main business cost and main business income mainly analyzes the difference between the main business cost / main business income in the current period and the same period last year. When the value is within [-0.3, 0.3], it is "in the normal fluctuation range". When the value is <-0.3, it indicates that "the downward trend is more significant compared with the previous year, which requires special attention". When the value is >0.3, it indicates that "the upward trend is more significant compared with the previous year, which requires special attention". When the value is == 0, it indicates that "it remains unchanged compared with the previous year";

[0129] When any of the conclusions in the above two parts is unreasonable, the conclusion “the risk of false cost reduction is high” is given; otherwise, “the risk of false cost reduction is low” is given.

[0130] The text interpretation template is as follows:

[0131] Combined with the analysis results of the financial whitewashing model, it is preliminarily determined that the risk of false cost reduction in the company's {2020 financial report} is {low}, mainly analyzed from the following two dimensions:

[0132] Trend of changes in accounts payable and prepaid expenses and main business income: As of the end of {December 2020}, the ratio of the year-on-year change in accounts payable and prepaid expenses to the year-on-year change in main business income of the company was {0.03}, while the ratio in the same period last year was {0.15}. The difference between the two periods (△Accounts payable and prepaid expenses / △Main business income) was {-0.11}, which is {within the normal fluctuation range}.

[0133] Changing trends of main business costs and main business income: As of the end of {December 2020}, the company's main business cost / main business income was {88.70%}, while the ratio was {89.12%} in the same period last year, a {decrease} of {0.42%} from the same period last year, {within the normal fluctuation range}.

[0134] The indicators involved will also be presented in tables and graphs for the past three years, as follows:

[0135]

[0136] 3. Analysis of inflated investment income: Regarding the investment income of the enterprise, this module analyzes the changing trends of long-term equity investment and main business income, and the changing trends of investment income and operating profit.

[0137] Calculation formula for the trend of long-term equity investment and main business income:

[0138]

[0139] The long-term equity investment and main business income change trend indicator is designed to evaluate the year-on-year change in long-term equity investment and the reasonableness of the year-on-year change in main business income. The reasonable threshold of this indicator is [-∞, 0.8]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating investment returns through fictitious growth period equity investment. When △ long-term equity investment / △ main business income > = 0.8, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in long-term equity investment or main business income. It is recommended to pay attention to the quality and rationality of long-term equity investment and main business income." Other situations are prompted: within the normal fluctuation range;

[0140] Calculation formula for investment income and operating profit trend:

[0141] The △Investment income / △Operating profit indicator is designed to assess the reasonableness of the year-on-year change in investment income and the year-on-year change in operating profit. The reasonable threshold of this indicator is [-∞, 0.5]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating investment income. When △Investment income / △Operating profit>=0.5, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in investment income or operating profit. It is recommended to pay attention to the quality and rationality of investment income and operating profit." In other cases, it is prompted: It is within the normal fluctuation range;

[0142] When any of the conclusions in the above two parts is unreasonable, the conclusion is "the risk of inflated investment returns is high", otherwise "the risk of inflated investment returns is low".

[0143] The text interpretation template is as follows:

[0144] Combined with the analysis results of the financial whitewashing model, it is preliminarily determined that the risk of inflated investment returns in the company's {2020 financial report} is {low}, mainly analyzed from the following two dimensions:

[0145] Changing trends of long-term equity investment and main business income: As of the end of {December 2020}, the year-on-year change in the company's long-term equity investment / year-on-year change in main business income was {0.24}, while the ratio in the same period last year was {0.22}. The difference between the two periods (△Long-term equity investment / △Main business income) was {0.03}, which is {within the normal fluctuation range}.

[0146] Trends in changes in investment income and operating profit: As of the end of {December 2020}, the year-on-year change in the company's investment income / operating profit was {-0.02}, while the ratio in the same period last year was {0.02}. The difference between the two periods (△investment income / △operating profit) was {-0.04}, which is {within the normal fluctuation range}.

[0147]

[0148] The indicators involved will also be presented in tables and graphs for the past three years, as follows:

[0149] The core value of this module lies in its comprehensiveness and depth. It interprets corporate financial fraud in a variety of ways and angles, providing users with accurate and reliable financial analysis results.

[0150] Step S9: Input the enterprise's financial data into the business risk model, and analyze the enterprise's business risk from five perspectives: solvency, profitability, operating capacity, growth capacity, and cash flow. Used to analyze the enterprise's business risk. Through in-depth analysis of five key aspects, combined with the numerical values ​​of various indicators of the same industry and scale in the "Enterprise Performance Evaluation Standard Value" formulated by the State-owned Assets Supervision and Administration Commission of the State Council, and the input financial report indicators for comparative analysis, this module provides a comprehensive perspective for the assessment of enterprise business risks.

[0151] These five areas include:

[0152] 1. Debt-paying ability analysis: First, the total debt of the enterprise and the comparison with the same period of the previous year will be displayed, and the values ​​of EBIT and EBITDA will be displayed. Then, the quick ratio, cash flow to debt ratio, debt-to-asset ratio, interest-bearing debt ratio, and interest coverage ratio will be analyzed. These indicators have specific numerical display and are compared with the excellent values, good values, average values, lower values, and poor values ​​corresponding to the same industry and the same scale in the "Enterprise Performance Evaluation Standard Values". Then, the industry level of the indicator is given as excellent, good, average, lower, or poor. Finally, the industry level of these indicators is combined to give the level of the company's debt-paying ability in the industry. The specific judgment logic is as follows:

[0153] When the interest-bearing debt ratio>performance standard average and interest coverage ratio<performance standard average and debt-to-asset ratio>performance standard average, the level is poor;

[0154] When the interest-bearing debt ratio>the lower value of the performance standard and the interest coverage ratio<the lower value of the performance standard and the debt-to-asset ratio>the lower value of the performance standard, the level is extremely poor;

[0155] When the interest-bearing debt ratio is less than or equal to the excellent performance standard, the level is good;

[0156] When the interest-bearing debt ratio <= the average performance standard and the debt-to-asset ratio <= the average performance standard, the level is good;

[0157] When the interest-bearing debt ratio <= the good performance standard value and the asset-liability ratio <= the good performance standard value, the level is good;

[0158] Other conditions are at average level.

[0159] The text interpretation template is as follows:

[0160] As of the end of {December 2020}, the company's total liabilities were {444.8691} million yuan, a {down} {1.89%} from the same period last year; EBIT was {58.3114} million yuan, and EBITDA was {58.3114} million yuan; the quick ratio was {80.08%}, which is at a {relatively low} level in the industry; the cash flow to debt ratio was {-3.68%}, which is at a {poor} level in the industry; the debt-to-asset ratio was {69.71%}, which is at a {relatively low} level in the industry; the interest-bearing debt ratio was {36.64%}, which is at an {average} level in the industry; and the interest coverage ratio was {2.76}, which is at a {relatively low} level in the industry.

[0161] Based on the above analysis, the company's debt repayment ability is at the {average} level in the industry.

[0162] This section will also display the above indicators in a chart. The example is as follows:

[0163]

[0164] 2. Profitability analysis: Through the analysis of key profitability indicators, users can understand the profitability and profitability of the company. This part will first display the total profit of the company and the comparison with the same period of the previous year, and then analyze the sales profit margin, gross profit margin, return on net assets, return on total assets, and EBITDA rate. These indicators have specific numerical display and are compared with the excellent values, good values, average values, lower values, and poor values ​​corresponding to the same industry and scale in the "Enterprise Performance Evaluation Standard Values", and then give the industry level of whether the indicator is excellent, good, average, lower, or poor. Finally, based on the industry level of these indicators, the profitability of the company is given at what level in the industry. The specific judgment logic is as follows:

[0165] When the sales profit margin is less than the lower value of the performance standard and the return on net assets is less than the average value of the performance standard and the return on total assets is less than the average value of the performance standard, the level is poor;

[0166] When the sales profit margin < the worse performance standard value and the return on net assets < the lower performance standard value and the return on total assets < the lower performance standard value, the level is extremely poor;

[0167] When the return on net assets>=the good value of the performance standard and the return on total assets>=the good value of the performance standard, the level is good;

[0168] When the return on net assets is greater than or equal to the good value of the performance standard and the sales profit margin is greater than or equal to the average value of the performance standard, the level is good;

[0169] When the return on total assets is greater than or equal to the good performance standard and the sales profit margin is greater than or equal to the average performance standard, the level is good;

[0170] When the sales profit margin>=lower value of the performance standard and the return on net assets>=average value of the performance standard and the return on total assets>=average value of the performance standard, the level is good;

[0171] When the sales profit margin>=the average value of the performance standard and the return on net assets>=the good value of the performance standard and the return on total assets>=the good value of the performance standard, the level is good;

[0172] Other conditions are at average level.

[0173] The text interpretation template is as follows:

[0174] As of the end of {December 2020}, the company's total profit was {37.181 million} yuan, {up} {15.84%} from the same period last year. The sales profit margin was {5.19%}, which was at a {good} level in the industry; the gross profit margin was {11.30%}, and the year-on-year change in gross profit margin was {0.42%}, {within the normal fluctuation range}; the return on net assets was {16.27%}, which was at an {excellent} level in the industry; the return on total assets was {9.38%}, which was at an excellent level in the industry; the EBITDA rate was {8.00%}, which was at a {good} level in the industry. Based on the above analysis, the profitability of the company is at a {good} level in the industry.

[0175] This section will also display the above indicators in a chart. The example is as follows:

[0176] Indicator name December 2018 December 2019 December 2020 Industry average Total owners' equity (ten thousand yuan) 11891.26 15223.54 19334.61 - Return on net assets (%) 19.65 17.94 16.27 3.20 Return on total assets (%) 10.34 8.68 9.38 1.50 EBITDA margin (%) 5.64 7.33 8.31 4.20 ... ... ... ... ...

[0177] 3. Operational Capacity Analysis: This part will focus on analyzing key indicators such as the company's asset turnover rate and inventory turnover rate to evaluate the company's operating efficiency. This part will analyze the company's total asset turnover rate, accounts receivable turnover rate, and inventory turnover rate at what level in the industry, as well as the year-on-year change compared with the same period last year, and analyze the fluctuations. Finally, based on the industry level of these indicators, the company's operating capacity is given at what level in the industry. The specific judgment logic is as follows:

[0178] When accounts receivable turnover < performance standard average and total asset turnover < performance standard average and inventory turnover < performance standard average, the level is poor;

[0179] When accounts receivable turnover times < performance standard lower value and total asset turnover times < performance standard lower value and inventory turnover times < performance standard lower value, the level is extremely poor;

[0180] When accounts receivable turnover times>=good performance standard value and total asset turnover times>=good performance standard value, the level is good;

[0181] When the accounts receivable turnover times>=good performance standard value and the inventory turnover times>=good performance standard value, the level is good;

[0182] When total asset turnover times>=good performance standard value and inventory turnover times>=good performance standard value, the level is good;

[0183] When accounts receivable turnover times>=performance standard average, total asset turnover times>=performance standard average, and inventory turnover times>=performance standard average, the level is good;

[0184] When accounts receivable turnover times>=good performance standard value and total asset turnover times>=good performance standard value and inventory turnover times>=good performance standard value, the level is good;

[0185] Other conditions are at average level.

[0186] The text interpretation template is as follows:

[0187] As of {the end of December 2020}, the company's total asset turnover rate was {1.13}, which is at a {good} level in the industry, and the year-on-year change in total asset turnover rate was {-0.06}, {within the normal fluctuation range}; the accounts receivable turnover rate was {2.53}, which is at the {average} level in the industry, and the year-on-year change in accounts receivable turnover rate was {0.05}, {within the normal fluctuation range}; the inventory turnover rate was {11.49}, which is at a {good} level in the industry, and the year-on-year change in inventory turnover rate was {-55.80}, {it is recommended to focus on analyzing the risk of reduced inventory utilization efficiency}.

[0188] Based on the above analysis, the company's operating capacity is at a relatively good level in the industry.

[0189] This section will also display the above indicators in a chart. The example is as follows:

[0190]

[0191] 4. Growth Capacity Analysis: This part will deeply explore the growth potential and trend of the enterprise, analyze the increase or decrease of the main business income, operating profit, and total assets of the enterprise compared with the same period last year, and compare and compare with the corresponding values ​​of the same industry and the same scale in the "Enterprise Performance Evaluation Standard Value", and give the industry level of the indicator as excellent, good, average, low or poor. Finally, based on the industry level of these indicators, the growth capacity of the enterprise is given at what level in the industry. The specific judgment logic is as follows:

[0192] When the growth rate of main business income is less than the average performance standard and the growth rate of operating profit is less than the average performance standard, the level is poor;

[0193] When the growth rate of main business income is less than the lower value of the performance standard and the growth rate of operating profit is less than the lower value of the performance standard, the level is extremely poor;

[0194] When the growth rate of main business income is greater than or equal to the excellent value of performance standard, the level is good;

[0195] When the operating profit growth rate is greater than or equal to the excellent value of the performance standard, the level is good;

[0196] When the growth rate of main business income is greater than or equal to the average value of performance standards and the growth rate of operating profit is greater than or equal to the average value of performance standards, the level is good;

[0197] When the growth rate of main business income is greater than or equal to the good value of performance standard and the growth rate of operating profit is greater than or equal to the good value of performance standard, the level is good;

[0198] Other conditions are at average level.

[0199] The text interpretation template is as follows:

[0200] As of the end of {December 2020}, the company's main business income was {701.8873} million yuan, {up} {11.74%} from the same period last year, and the growth rate of main business income was at a {good} level in the industry; operating profit was {36.4425} million yuan, {up} {13.02%} from the same period last year, and the growth rate of operating profit was at the {average} level in the industry; total assets were {638.2152} million yuan, {up} {5.37%} from the same period last year; the technology investment ratio was {0.00%}, which was at a {poor} level in the industry.

[0201] Based on the above analysis, the company's growth capability is at a relatively good level in the industry.

[0202] This section will also display the above indicators in a chart. The example is as follows:

[0203] Indicator name December 2018 December 2019 December 2020 Industry average Growth rate of main business income (%) 19.65 -16.75 11.74 2.70 Operating profit growth rate (%) 3.16 19.04 13.02 -150.00 Technology input ratio (%) 0.00 0.00 0.00 0.90 ... ... ... ... ...

[0204] 5. Cash flow analysis: In view of the cash flow status of the enterprise, this module analyzes the indicators including the net cash flow generated by the enterprise's operating activities, cash inflows from operating activities, cash outflows from operating activities, net cash flow generated by investment activities, cash inflows from investment activities, cash outflows from investment activities, net cash flow generated by financing activities, cash inflows from financing activities, and cash outflows from financing activities in the current period compared with the same period last year, analyzes the cash collection ratio indicator, and judges the ability of converting operating income into cash. >=1.27 indicates good ability, >=0.67 indicates average, and other situations are poor. Finally, the conclusion is given. If the net cash flow generated by the operating activities of the current period is greater than 0 compared with the net cash flow generated by the operating activities of the same period last year, it indicates that "the cash flow of the enterprise has been sufficient in the past two years". If the net cash flow generated by the operating activities of the current period is less than 0 compared with the net cash flow generated by the operating activities of the same period last year, it indicates that "the cash flow of the enterprise has been relatively tight in the past two years, and it is recommended to continue to pay attention to the performance risk of the enterprise". Other situations indicate that "the cash flow of the enterprise has been average in the past two years".

[0205] The text interpretation template is as follows:

[0206] As of the end of December 2020, the company's net cash flow from operating activities was RMB 16.3915 million, a decrease of 483.53% from the same period last year; cash inflow from operating activities was RMB 152.01473 million, an increase of 27.88% from the same period last year; cash outflow from operating activities was RMB 153.65389 million, an increase of 27.88% from the same period last year; net cash flow from investment activities was RMB 21.5143 million, an increase of 25.94% from the same period last year; cash inflow from investment activities was RMB 29.030 million, an increase of 27.88% from the same period last year. During the same period, it increased by 190.15%, and the cash outflow from investment activities was 24.4173 million yuan, which was 18.75% lower than that of the same period last year. The net cash flow from financing activities was 55.5067 million yuan, which was 772.47% higher than that of the same period last year. The cash inflow from financing activities was 192 million yuan, which was 121.50% higher than that of the same period last year. The cash outflow from financing activities was 136.4933 million yuan, which was 43.78% higher than that of the same period last year. The cash collection ratio was 104.00%, and the ability to convert operating income into cash was relatively good.

[0207] Based on the above analysis, the company's cash flow in the past two years has been relatively tight, and it is recommended to continue to pay attention to the company's performance risks.

[0208] This section will also display the above indicators in a chart. The example is as follows:

[0209]

[0210] The comprehensive recommendations of this module are based on in-depth analysis and comparison of various key indicators. Through a combination of text description and chart display, it provides users with a comprehensive assessment and recommendation of corporate operating risks.

[0211] The above description is only the preferred embodiment of the present invention, which does not limit the protection scope of the present invention. For those skilled in the art, the present invention can be modified and varied in various ways. Within the spirit and principle of the present invention, any change, modification, replacement, integration and parameter change of these embodiments by conventional substitution or capable of achieving the same function without departing from the principle and spirit of the present invention shall fall within the protection scope of the present invention.

Claims

1. A financial risk identification method based on machine learning, characterized in that: include: Step S1, collect the test data set required by the model; Step S2, divide the test data set into four categories: classic financial indicators, empirical financial indicators, whitewashing indicators, and Themis indicators, and construct 280 derived financial indicators; Step S3, analyzing the importance of the derived financial indicators, screening out the derived financial indicators with high feature importance ranking and information value IV exceeding 0.02, and ensuring that their trends are interpretable; Step S4: input the screened derivative financial indicator data and the corresponding financial whitewashing labels into the XGBoost model for training, and adjust the training parameters of XGBoost to obtain the financial whitewashing model; input the screened derivative financial indicator data and the corresponding operating risk labels into the XGBoost model for training, and adjust the training parameters of XGBoost to obtain the operating risk model; Step S5: construct a cross-checking relationship module by presetting the cross-checking relationship between multiple financial basic indicators; when the financial report basic indicators input by the enterprise do not meet the preset cross-checking relationship, the system will display and output corresponding descriptive text; Step S6, establish an early warning indicator module and set a threshold; when the enterprise inputs financial data and industry information, the early warning indicator module calculates the enterprise early warning indicator value and compares it with the threshold. If the threshold is triggered, the early warning indicator module will output the triggered early warning indicator, specific data, risk description and matters of concern; Step S7: Establish a balance sheet display module to obtain the company's debt situation, and use text description and chart display to analyze key financial indicators and analyze the asset composition structure in detail; Step S8: Input the enterprise's financial data into the financial whitewashing model to analyze the enterprise's financial whitewashing from three perspectives: inflated revenue, reduced costs, and inflated investment returns; Step S9: Input the enterprise's financial data into the business risk model, and analyze the enterprise's business risk from five perspectives: debt repayment ability, profitability, operating ability, growth ability, and cash flow.

2. The method for identifying financial risks based on machine learning according to claim 1, characterized in that: The test data set includes: financial samples and business samples; The financial samples include: good financial samples and bad financial samples; The financial bad samples include: collecting administrative penalty data of listed companies in the past ten years from the website of the China Securities Regulatory Commission, and then parsing the penalty information texts, defining listed companies that have been punished for financial whitewashing and false financial reports as bad customers; for these bad customers, using API data sources to collect their reported current financial reports as bad samples, and paying special attention to the fact that these data are not revised financial reports; The good financial samples include: defining listed companies that have not been punished by the China Securities Regulatory Commission for financial fraud as good customers; for these good customers, using API data sources to collect their financial reports for the past five years as good samples; The business samples include: good business samples and bad business samples; The good operating samples include: collecting key information such as the financial report data, enterprise size, industry, and five-level post-loan classification of the company's customers; screening out customers who have been in a state of concern or bad in the past four years as bad customers, and selecting the financial reports of these bad customers in the six months before they entered the state of concern or bad as bad samples; The bad operating samples include: customers who have had credit in the bank for nearly four years and whose five-level classification is normal are defined as good customers, and the financial statements of these good customers are randomly selected as good samples.

3. The method for identifying financial risks based on machine learning according to claim 1, characterized in that: In step S5, there are 9 cross-references, including: Monetary funds at the end of the period ≠ monetary funds at the beginning of the period + net increase in cash and cash equivalents at the end of the period; Undistributed profit at the end of the period > Undistributed profit at the beginning of the period + Net profit at the end of the period; Total assets ≠ total liabilities + total owner's equity; Net cash flow from operating activities ≠ Subtotal of cash inflow from operating activities - Subtotal of cash outflow from operating activities; Net cash flow from investing activities ≠ Subtotal of cash inflows from investing activities - Subtotal of cash outflows from investing activities; Net cash flow from financing activities ≠ Subtotal of cash inflows from financing activities - Subtotal of cash outflows from financing activities; Net cash increase ≠ net cash flow from operating activities + net cash flow from investing activities + net cash flow from financing activities; Total comprehensive income ≠ net profit + other comprehensive income; Net profit ≠ total profit - income tax expense.

4. The method for identifying financial risks based on machine learning according to claim 1, characterized in that: The step S6 specifically includes: Step S61: Based on the sample data of each industry, draw an IV chart of indicators applicable to nine industries, including electricity, heat, gas and water production and supply, real estate, construction, transportation, warehousing and postal services, wholesale and retail, water conservancy, environment and public facilities management, information transmission, software and information technology services, manufacturing, leasing and business services, as well as all industries; divide the derived financial indicators into boxes, and draw a corresponding IV chart with each group as the horizontal axis and the number and bad sample rate of each group as the double vertical axes; Step S62: Combine the IV graph trend to select early warning indicators with early warning significance; Step S63: Determine the optimal threshold of the warning indicator based on the selected warning indicators by combining the precision rate and the alarm trigger rate; Precision rate = the number of bad customers triggered by a single warning indicator / the number of customers triggered by a single warning indicator. The precision rate evaluates whether the warning indicator can effectively warn, which requires more than or equal to twice the number of bad samples in the industry; Trigger rate = number of customers triggered by a single warning indicator / total number of customers. The trigger rate evaluates the alarm ratio of the warning indicator and should not be too large, less than or equal to 0.1; Step S64: Determine the warning level according to the precision of the customer whose indicator triggers the alarm; Step S65: Output the risk description and concerns of the warning indicators.

5. The method for identifying financial risks based on machine learning according to claim 1, characterized in that: In step S8, the analysis of inflated income includes: analyzing from two dimensions, the trend of changes in accounts receivable and main business income, the trend of changes in non-operating income and total operating income, and comparing the difference between the current value and the same period last year; if the difference fluctuates too much and exceeds the preset standard threshold, the financial whitewashing model will issue a risk warning; The trend of changes in accounts receivable and main business income mainly analyzes the Δ accounts receivable / Δ main business income indicator; The Δ accounts receivable / Δ main business income indicator is used to evaluate the reasonableness of the year-on-year change in accounts receivable and the year-on-year change in main business income. The reasonable threshold of this indicator is [-∞,1.5]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating the main business income by inflating accounts receivable. When Δ accounts receivable / Δ main business income>=1.5, it is prompted: "Beyond the normal fluctuation range, there is unreasonable situation in the change of accounts receivable or main business income, it is recommended to pay attention to the quality and rationality of accounts receivable and main business income", and other situations are prompted: "Within the normal fluctuation range"; The trend of changes in non-operating income and total operating income is analyzed by analyzing the growth rate of non-operating income and the proportion of non-operating income to total operating income. If the growth rate of non-operating income is greater than 1 and the proportion of non-operating income to total operating income is greater than or equal to 0.05, it is prompted that "non-operating income grows rapidly and accounts for a high proportion of total operating income. There are unreasonable situations in the changes of non-operating income and total operating income. It is recommended to pay attention to the quality and rationality of non-operating income and total operating income." In other cases, it is prompted that "non-operating income and total operating income show a normal trend of change"; When any of the conclusions in the above two parts is unreasonable, the conclusion "the risk of inflated income is high" is given; otherwise, "the risk of inflated income is low".

6. The method for identifying financial risks based on machine learning according to claim 1, characterized in that: In step S8, the virtual cost reduction analysis includes: analyzing the key indicators of the change trend of accounts payable & prepaid and main business income, the change trend of main business cost and main business income, and performing text interpretation and chart display; The main analysis of the trend of changes in accounts payable & prepaid and main business income is △ accounts payable & prepaid / △ main business income indicators, The trend indicator of accounts payable & prepaid and main business income is intended to evaluate the reasonableness of the year-on-year change in accounts payable & prepaid and the year-on-year change in main business income. The reasonable threshold of this indicator is [-2.5,∞]; the greater the deviation between the indicator value and the reasonable threshold range, the higher the risk of falsely reducing costs by falsely reducing accounts payable & prepaid; when △ accounts payable & prepaid / △ main business income <= -2.5, it is prompted: "Beyond the normal fluctuation range, there is unreasonable situation in the change of accounts payable & prepaid or main business income, it is recommended to pay attention to the quality and rationality of accounts payable & prepaid and main business income" other situations are prompted: within the normal fluctuation range; The trend of changes in main business cost and main business income mainly analyzes the difference between the main business cost / main business income in the current period and the same period last year. When the value is within [-0.3, 0.3], it is "in the normal fluctuation range", when the value is <-0.3, it indicates "the downward trend is more significant compared with the previous year, which requires special attention", when the value is >0.3, it indicates "the upward trend is more significant compared with the previous year, which requires special attention", when the value == 0, it indicates "remains unchanged compared with the previous year"; When any of the conclusions in the above two parts is unreasonable, the conclusion "the risk of false cost reduction is high" is given; otherwise, "the risk of false cost reduction is low".

7. The method for identifying financial risks based on machine learning according to claim 1, characterized in that: In step S8, the analysis of inflated investment returns includes: Analyze the changing trends of long-term equity investment and main business income, and the changing trends of investment income and operating profit; Calculation formula for the trend of long-term equity investment and main business income: The long-term equity investment and main business income change trend indicator is designed to evaluate the year-on-year change in long-term equity investment and the reasonableness of the year-on-year change in main business income. The reasonable threshold of this indicator is [-∞, 0.8]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating investment returns through fictitious growth period equity investment. When △ long-term equity investment / △ main business income > = 0.8, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in long-term equity investment or main business income. It is recommended to pay attention to the quality and rationality of long-term equity investment and main business income." Other situations are prompted: within the normal fluctuation range; Calculation formula for investment income and operating profit trend: The investment income / △operating profit indicator is designed to assess the reasonableness of the year-on-year change in investment income and the year-on-year change in operating profit. The reasonable threshold of this indicator is [-∞, 0.5]. The greater the deviation between the indicator value and the reasonable threshold interval, the higher the risk of inflating investment income. When △investment income / △operating profit>=0.5, it is prompted: "Beyond the normal fluctuation range, there are unreasonable changes in investment income or operating profit. It is recommended to pay attention to the quality and rationality of investment income and operating profit." In other cases, it is prompted: within the normal fluctuation range; When any of the conclusions in the above two parts is unreasonable, the conclusion is "the risk of inflated investment returns is high", otherwise "the risk of inflated investment returns is low".

Images