A Single Dimension Strategy Automatic Mining Method
Through the single-dimensional strategy analysis in the financial risk control field, the threshold calculation and generalization analysis of the sample data of strategy are automatically processed, and the problem of cumbersome variable binning and analysis steps is solved, and efficient and automated single-dimensional strategy analysis is achieved.
Patent Information
- Application Number
- CN202111274003.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-29
- Publication Date
- 2025-02-18
- Estimated Expiration
- 2041-10-29
AI Technical Summary
In the single-dimensional strategy analysis in the field of financial risk control, the variable binning methods are numerous and cumbersome, and artificially determining the threshold leads to low analysis efficiency and includes subjective factors; at the same time, the single-dimensional strategy analysis steps are cumbersome and time-consuming, making it difficult to achieve automation and batching analyze.
It provides a single-dimensional strategy automatic mining method, which can automatically output standardized analysis reports by obtaining policy sample data of different time spans, performing threshold calculations and generalization analysis, and automatically output standardized analysis reports to realize the automation and batching of strategy analysis.
The efficiency of single-dimensional strategy analysis is improved, the influence of human subjective factors is avoided, the analysis process is fully automated, and standardized analysis reports are output.
Smart Images
Figure CN113988191B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a single-dimensional strategy automatic mining method. Background Art
[0002] At present, in the field of financial risk control, when conducting single-dimensional strategy mining, it is generally necessary to analyze thousands of variables, and select several variables with good effects and business explanatory value based on the analysis results to make single-dimensional strategies for risk interception. There are many difficulties in conducting single-dimensional strategy analysis, as follows:
[0003] (1) When conducting single-dimensional strategy analysis, there are many ways to bin variables. After binning, the Odds_Ratio risk multiples and touch conditions of different bins of each variable are often manually checked, and the threshold of the variable is manually determined. The analysis process is slow and contains many subjective factors.
[0004] (2) Single-dimensional strategy analysis includes various analysis steps, and the complete analysis steps are cumbersome and time-consuming.
[0005] (3) If single-dimensional strategy analysis is performed on thousands of variables at the same time, the analysis efficiency will be low due to the large number of analysis variables. Summary of the invention
[0006] The main purpose of the present invention is to provide a single-dimensional strategy automatic mining method, aiming to achieve single-dimensional strategy optimization, automation, and batch analysis, while automatically outputting standardized analysis reports to improve the efficiency of single-dimensional strategy analysis.
[0007] The basic solution provided by the present invention is:
[0008] A single-dimensional strategy automatic mining method, comprising:
[0009] Obtain strategy sample data of different time spans required for single-dimensional strategy analysis;
[0010] Performing threshold calculation on the strategy sample data and outputting a strategy threshold calculation analysis report;
[0011] According to the strategy threshold calculation analysis report, the strategy samples that meet the preset conditions and their corresponding optimal thresholds are screened for generalization analysis to output a first generalization analysis report;
[0012] According to the first generalization analysis report, policy samples whose performance meets expectations are screened for online application, and multiple policy samples whose performance meets expectations are merged and generalization analysis is performed again to output a second generalization analysis report.
[0013] The principle of the basic scheme of the present invention is:
[0014] In this solution, the single-dimensional strategy automatic mining method is applied to the field of financial risk control. When conducting single-dimensional strategy mining, it is generally necessary to analyze thousands of strategy variables. Based on the analysis results of the strategy variables, several strategy variables with good effects and business interpretability are selected to form a single-dimensional strategy, and the single-dimensional strategy is put online to intercept risks in the field of financial risk control. Specifically, firstly, the strategy sample data of different time spans required for the single-dimensional strategy analysis are obtained, the threshold is calculated for each strategy sample, and a strategy threshold calculation analysis report with each strategy sample is output, which can flexibly screen the required strategy variables and improve convenience, while avoiding the problem of manually determining the variable threshold, which leads to a slow analysis process and contains many subjective factors; then, according to the strategy threshold calculation analysis report, the strategy samples that meet the preset conditions and their corresponding optimal thresholds are screened for generalization analysis, so as to automatically output the first generalization analysis report, automate the entire strategy analysis process, automatically find the strategy variables that meet the preset conditions and give online suggestions, greatly improving the efficiency of strategy analysis; finally, according to the first generalization analysis report, the strategy samples whose performance meets the expectations are screened for online application, and multiple strategy samples whose performance meets the expectations are merged, and generalization analysis is performed again to automatically output the second generalization analysis report, so as to evaluate what kind of effect can be achieved after multiple strategies are put online.
[0015] That is, when conducting single-dimensional strategy analysis, this solution optimizes the strategy analysis binning method based on business logic and realizes full automation of the analysis process, avoiding the problem of manually determining variable thresholds, which leads to a slow analysis process and contains many subjective factors, and automatically outputs standardized analysis reports, which can greatly improve the analysis efficiency of strategy variables, especially when analyzing a large number of strategy variables.
[0016] Furthermore, the strategy sample data includes independent variable data and dependent variable data.
[0017] In this solution, each variable data in the strategy sample data is used to facilitate threshold calculation for each strategy sample in the strategy sample data.
[0018] Furthermore, the step of obtaining strategy sample data of different time spans required for single-dimensional strategy analysis includes:
[0019] Obtaining an independent variable data dictionary; the independent variable data dictionary includes the English name of the strategy sample, the Chinese name of the strategy sample and the strategy sample type;
[0020] According to the obtained independent variable data dictionary, the Chinese name of the strategy sample is matched for each independent variable data during strategy sample analysis.
[0021] In this solution, by obtaining the independent variable data dictionary, the Chinese name of the strategy sample is conveniently matched to each independent variable data, so as to quickly prepare the parameters required for single-dimensional strategy analysis and improve the efficiency of single-dimensional strategy analysis.
[0022] Further, the steps of performing threshold calculation on the policy sample data and outputting a policy threshold calculation analysis report include:
[0023] Performing descriptive statistical analysis on the strategy sample data and performing a primary screening after the descriptive statistical analysis, and marking whether the primary screening is performed as a first label;
[0024] The non-missing samples in the strategy sample data filtered by the first label are sorted from small to large according to the indicator values, and then subdivided and roughly divided, and the sample size and corresponding indicators corresponding to different bins are output;
[0025] Among the samples in the preset range of non-missing sample bins, the threshold with the largest local indicator sample in the preset touch rate range is selected as the optimal threshold;
[0026] Determine whether the local indicator sample and the global indicator sample meet the preset conditions, and mark the policy sample data that meets the preset conditions as the second label;
[0027] Based on descriptive statistical analysis, optimal binning, screening of variables with good performance based on optimal binning and threshold output strategy threshold calculation analysis report.
[0028] In this solution, the strategy sample data is subjected to descriptive statistical analysis, initial screening, variable binning, optimal variable threshold determination, variable performance analysis and variable re-screening to achieve threshold calculation for each strategy sample in the strategy sample data, and output a strategy threshold calculation analysis report, thereby achieving full automation of the analysis process, avoiding the problem of manual determination of variable thresholds, which leads to a slow analysis process and contains many subjective factors, and further improving the analysis efficiency of strategy variables.
[0029] Further, the step of judging whether the local indicator sample and the global indicator sample meet the preset conditions and marking the policy sample data meeting the preset conditions as the second label includes:
[0030] According to reducing the relevance of the screening variable, the policy sample data that meets the preset condition in the policy sample data marked as the second label is marked as the third label.
[0031] In this solution, based on reducing the correlation of the screening variables, the strategy sample data that meets the preset conditions in the local indicator samples and global indicator samples marked as the second label are marked as the third label.
[0032] Furthermore, the threshold calculation analysis report based on descriptive statistical analysis, optimal binning, screening of variables with good performance based on optimal binning, and threshold output strategy is specifically as follows:
[0033] Based on descriptive statistical analysis, optimal binning, and screening of variables and thresholds with good performance based on optimal binning, the strategy threshold measurement and analysis report in Excel format is automatically output through Python.
[0034] In this solution, Python is used to more quickly and accurately capture various parameters based on variable threshold calculation results, thereby forming a strategy threshold calculation analysis report.
[0035] Further, the step of screening the policy samples that meet the preset conditions and their corresponding optimal thresholds for generalization analysis according to the threshold measurement and analysis report to output a first generalization analysis report includes:
[0036] Filtering strategy samples with preset effectiveness according to the strategy threshold measurement and analysis report;
[0037] A generalization analysis is performed on the optimal threshold of the strategy sample with the preset effectiveness screened out according to the threshold generalization template to output a first generalization analysis report.
[0038] In this scheme, the strategy samples with preset effectiveness are further screened according to the strategy threshold measurement and analysis report, and then the optimal threshold of the screened strategy samples is generalized and analyzed based on the threshold generalization template to output a standard generalization analysis report more accurately and reliably.
[0039] Furthermore, the generalization analysis of the optimal threshold of the strategy sample with the preset effectiveness is performed according to the threshold generalization template as follows:
[0040] Analyze and filter out the optimal thresholds of strategy samples with preset performance, touch rate and performance in different time periods, and automatically output indicators on whether to recommend strategy launch.
[0041] In this solution, by analyzing the touch rate and effectiveness of strategy samples in different time periods, more accurate and reliable suggestions can be output to recommend whether the strategy should be launched.
[0042] Furthermore, the first generalization analysis report includes generalization description, rule generalization and touch simulation analysis.
[0043] In this solution, the data specifically included in the variables through generalization analysis are directly displayed in the first generalization analysis report, so as to facilitate the analysis of a large number of variables during the strategy sample analysis and improve the efficiency of the analysis process.
[0044] Further, the steps of screening strategy samples whose performance meets expectations according to the first generalization analysis report for online application, merging multiple strategy samples whose performance meets expectations, and performing generalization analysis again to output a second generalization analysis report include:
[0045] According to the first generalization analysis report, a preset number of strategy samples to be launched whose effectiveness meets expectations are selected for online launch;
[0046] The selected strategy samples whose performance meets expectations are merged into strategy sample data with touch threshold values of 0 and 1;
[0047] Perform variable threshold generalization analysis according to the threshold generalization template and automatically output the second generalization analysis report.
[0048] In this scheme, a preset number of strategies to be launched are screened out through the analysis data in the first generalized analysis report, and the preset number of strategies to be launched are merged to perform generalized analysis again according to the threshold generalization template, so as to simulate and analyze indicators such as strategy effectiveness and strategy touch status after the strategies to be launched are launched at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 A schematic diagram of the internal structure of a terminal device involved in an embodiment of the present invention;
[0050] Figure 2 The figure is a flow chart of an embodiment of a method for automatic mining of single-dimensional strategies according to the present invention. DETAILED DESCRIPTION
[0051] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0052] The following is further described in detail through specific implementation methods:
[0053] Reference numerals in the accompanying drawings of the specification include: processor 1001 , communication bus 1002 , user interface 1003 , network interface 1004 , and memory 1005 .
[0054] like Figure 1 As shown, it is a schematic diagram of the internal structure of the terminal device involved in the embodiment of the present invention.
[0055] It should be noted that Figure 1 That is, it is a schematic diagram of the structure of the hardware operating environment of the terminal device. The terminal device in the embodiment of the present invention can be a PC, a portable computer or other terminal devices.
[0056] like Figure 1As shown, the terminal device may include: a processor 1001, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between the processor 1001, the user interface 1003, the network interface 1004, and the memory 1005. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), a handwriting board, a stylus, etc., and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface (such as an RJ45 interface) and a wireless interface (such as a WIFI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.
[0057] Those skilled in the art will understand that Figure 1 The terminal device structure in the figure does not constitute a limitation on the terminal device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.
[0058] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a distributed task processing program. Among them, the operating system is a program that manages and controls the hardware and software resources of the sample terminal device, and supports the operation of the distributed task processing program and other software or programs.
[0059] exist Figure 1 In the terminal device shown, the user interface 1003 is mainly used to communicate data with each terminal; the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; and the processor 1001 can be used to call the single-dimensional strategy automatic mining program in the memory 1005 and execute the following steps: Figure 2 The following operations are shown:
[0060] Step S10, obtaining strategy sample data of different time spans required for single-dimensional strategy analysis;
[0061] Step S20, performing threshold calculation on the policy sample data, and outputting a policy threshold calculation analysis report;
[0062] Step S30, screening the policy samples that meet the preset conditions and their corresponding optimal thresholds according to the policy threshold measurement and analysis report for generalization analysis to output a first generalization analysis report;
[0063] Step S40, screening policy samples whose performance meets expectations according to the first generalization analysis report for online application, merging multiple policy samples whose performance meets expectations, performing generalization analysis again, and outputting a second generalization analysis report.
[0064] In this related technology, when performing single-dimensional strategy analysis, there are many methods for variable binning, and after binning the variables, it is often necessary to manually check the risk multiples and touch conditions of different bins after binning each variable, and manually determine the threshold of the variable. The analysis process is slow and contains many subjective factors. Based on this, the problem is: whether the variables can be flexibly binned based on the efficiency of the variables, and the optimal threshold of the variables can be determined at the same time. Secondly, single-dimensional strategy analysis includes various analysis steps, and the complete analysis steps are relatively cumbersome and time-consuming; based on this, the problem is: whether the various steps in the single-dimensional strategy analysis can be integrated, the strategy analysis can be automatically performed, and a standardized strategy analysis report can be output. In addition, if single-dimensional strategy analysis is performed on thousands of variables at the same time, the analysis efficiency is low due to the large number of analysis variables; the problem is: whether the single-dimensional strategy analysis of all variables can be automatically completed in batches, and at the same time, the variables with good effects can be automatically found, the variable online threshold can be determined, and the standardized strategy analysis report can be automatically output, which greatly improves the analysis efficiency.
[0065] Based on the problems existing in the relevant technology, in this solution, firstly, the policy sample data of different time spans required for the single-dimensional policy analysis are obtained, the threshold is calculated for each policy sample, and a policy threshold calculation analysis report with each policy sample is output, which can flexibly screen the required policy variables and improve convenience, while avoiding the problem of manually determining the variable threshold, which leads to a slow analysis process and contains many subjective factors; then, according to the policy threshold calculation analysis report, the policy samples that meet the preset conditions and their corresponding optimal thresholds are screened for generalization analysis, so as to automatically output the first generalization analysis report, automate the entire policy analysis process, automatically find the policy variables that meet the preset conditions and give online suggestions, greatly improving the efficiency of policy analysis; finally, according to the first generalization analysis report, the policy samples whose performance meets the expectations are screened for online application, and multiple policy samples whose performance meets the expectations are merged, and generalization analysis is performed again to automatically output the second generalization analysis report, so as to evaluate what kind of effect can be achieved after multiple strategies are put online. It should be noted that the single-dimensional strategy automatic mining method is applied in the field of financial risk control. When conducting single-dimensional strategy mining, it is generally necessary to analyze thousands of strategy variables. Based on the analysis results of the strategy variables, several strategy variables with good effects and business interpretability are screened to form a single-dimensional strategy, and the single-dimensional strategy is put online to intercept risks in the field of financial risk control.
[0066] That is, when conducting single-dimensional strategy analysis, this solution optimizes the strategy analysis binning method based on business logic and realizes full automation of the analysis process, avoiding the problem of manually determining variable thresholds, which leads to a slow analysis process and contains many subjective factors, and automatically outputs standardized analysis reports, which can greatly improve the analysis efficiency of strategy variables, especially when analyzing a large number of strategy variables.
[0067] Based on the above embodiment, the step of obtaining strategy sample data of different time spans required for single-dimensional strategy analysis includes: obtaining an independent variable data dictionary; the independent variable data dictionary includes the strategy sample English name, the strategy sample Chinese name and the strategy sample type; according to the obtained independent variable data dictionary, matching the strategy sample Chinese name for each independent variable data during strategy sample analysis. It should be noted that the strategy sample data includes independent variable data and dependent variable data. It should be noted that the strategy sample is the independent variable, and the strategy samples with good effects are screened from tens of thousands of strategy sample data to make a single-dimensional strategy online for risk interception.
[0068] In one embodiment, the step of performing threshold calculation on the policy sample data and outputting a policy threshold calculation analysis report specifically includes:
[0069] Step S21: perform descriptive statistical analysis on the strategy sample data and perform initial screening after the descriptive statistical analysis, and mark whether to perform initial screening as the first label; specifically, the analysis of a single dimension includes automatically matching the Chinese name of the variable based on the data dictionary, sample size, missing value, missing rate, bad rate Badrate, variable value with the largest proportion of a single value, sample size with the largest proportion of a single value, largest proportion of a single value, variable value with the second largest proportion of a single value, sample size with the second largest proportion of a single value, second largest proportion of a single value, variable value with the third largest proportion of a single value, sample size with the third largest proportion of a single value, third largest proportion of a single value, total sample size of the top two largest proportions of a single value, sum of the top two largest proportions of a single value, total sample size of the top three largest proportions of a single value, sum of the top three largest proportions of a single value, number of variable values containing missing values, number of variable values excluding missing values, minimum value, maximum value, mean value, quantile, standard deviation, coefficient of dispersion, and marking the variable with a first label (Y / N), i.e., the initial screening label, based on the variable missing rate and single value proportion.
[0070] It should be noted that samples with mature target field performance in a certain period of time in the strategy sample data (mature samples are selected based on the label of whether the target field in the sample is mature) are selected as training sets for threshold measurement, and the strategy threshold measurement and analysis report is automatically output. The strategy threshold measurement and analysis report mainly includes descriptive statistical analysis, optimal binning of variables, variable screening based on optimal binning, etc. It is understandable that a certain period of time can be, but is not limited to, January-February 2021, May-July 2021, etc.
[0071] Based on the results of descriptive statistical analysis, the variables are initially screened. Variables with missing rates below a certain threshold enter the next round of analysis, and whether or not to conduct the initial screening is marked as the first label. Only variables with low missing rates are worthwhile for the next step of analysis. If the missing rate is very high, there is no analysis value.
[0072] Step S22: Variable binning, that is, sorting the non-missing samples in the strategy sample data filtered by the first label according to the index value from small to large, and then performing subdivision and rough division, and outputting the sample size and corresponding indicators corresponding to different bins; specifically: (1) sorting the non-missing samples according to the index value from small to large; (2) subdividing the N% before and after the index value of the non-missing samples, and the proportion of each group shall not exceed K%, and the number of samples in each group shall not be less than n; (3) performing rough division on the middle (1-2*N%) of the index value of the non-missing samples, and the proportion of each group shall not exceed S%, where the value of S% needs to be less than or equal to (1-2*N%), for example, it can be 10%. Here, steps (1)-(3) are for the processing of non-missing samples, except for the missing samples, the remaining missing values are boxed separately. It should be noted that the binning indicators include the Chinese name of the binning variable, the binning range, the sample size of each box, the sample proportion, the good sample size, the good sample proportion, the bad sample size, the bad sample proportion, the WOE value, the IV value, the local indicator sample (Odds_Ratio) positive value, the local indicator sample (Odds_Ratio) negative value, the risk multiple value, etc.; N% is the ratio of non-missing samples, which can be but not limited to 5%, 6%, etc.; K% can be but not limited to 1%, 2%, etc.; n can be but not limited to 30, 35, etc., and n must at least satisfy the law of large numbers.
[0073] Step S23: Determine the optimal threshold value of the variable, that is, in the preset range samples of the non-missing sample bin, select the threshold value with the preset touch rate range and the largest local indicator sample as the best threshold value; specifically: in the N% strategy samples before and after the non-missing value bin, select the threshold value with the touch rate between (0.05%, N%) and the largest local indicator sample (both the local indicator samples calculated from the positive and negative sample angles are analyzed) as the optimal threshold value.
[0074] Step S24: variable performance analysis and variable re-screening, that is, judging whether the local indicator samples and the global indicator samples meet the preset conditions, and marking the strategy sample data that meet the preset conditions as the second label; according to reducing the correlation of the screening variables, the strategy sample data that meet the preset conditions in the strategy sample data marked as the second label are marked as the third label. Specifically, the Odds_Ratio (calculated based on non-missing value samples) and risk multiple (calculated based on the full number of samples) of the samples touching the optimal threshold are calculated, and it is determined whether the Odds_Ratio and risk multiple of the samples touching the optimal threshold are greater than or equal to 2. Based on the determination result, a second label (Y / N) is assigned to the strategy variable; the same variable may exist in multiple time windows (such as the variable of the number of multiple lending institutions: the variable of the number of multiple lending institutions in the past 1 month, 3 months, 6 months, and 9 months) and is labeled with a second label of Y: (1) If the same variable is labeled with a second label of Y in less than or equal to 2 time windows, the variable is labeled with a third label (Y); (2) If the same variable is labeled with a second label of Y in more than 2 time windows, in order to avoid selecting too many related variables for strategy generalization, only the variable with the highest risk multiple and the time span closest to the observation point is labeled with the third label (Y); (3) For other variables, the third label is N.
[0075] In this embodiment, the preset condition is that after the non-missing value samples are optimally binned, the Odds_Ratio corresponding to the head or tail bin is not less than the preset Odds_Ratio value, and after the missing values are taken into account, the risk multiple corresponding to the head or tail bin is not less than the preset risk multiple value. It should be noted that the preset Odds_Ratio value can be, but is not limited to, 3, etc., and the missing values are not taken into account when calculating the Odds_Ratio, so the calculated indicator is a local indicator; the preset risk multiple value can be, but is not limited to, 3.5, etc., and both missing samples and non-missing samples are taken into account when calculating the risk multiple, which is a global indicator.
[0076] In order to reduce the correlation of the screening variables, for the strategy samples screened out by the second label, if some of these strategy samples are strongly linearly correlated, it is not necessary to make all of them online, and no more than two strategy samples with strong correlation can be online. Through linear analysis of these strategy samples, some strategy samples with strong linear correlation are eliminated, and the strategy samples screened out by the second label are further screened and marked with the third label; the strategy samples screened by the third label are the strategy samples that need to be generalized for the first time.
[0077] Step S25: Automatically output the strategy threshold measurement and analysis report (excel), that is, output the strategy threshold measurement and analysis report based on descriptive statistical analysis, optimal binning, and screening of variables and thresholds with good performance based on optimal binning. The strategy threshold measurement and analysis report specifically includes: variable screening summary, describing the steps and methods of single-dimensional strategy threshold measurement and analysis, the number of statistical analysis variables, the number of variables screened by the first label, the second label, and the third label, etc.; descriptive statistical analysis and screening; variable binning; variable performance analysis and screening. Based on the third label field of the form, you can check which variables are finally screened for the next step of threshold generalization analysis. It should be noted that the strategy threshold measurement and analysis report is an analysis report automatically output in excel format through Python.
[0078] In one embodiment, the step of screening the policy samples that meet the preset conditions and their corresponding optimal thresholds for generalization analysis according to the policy threshold measurement and analysis report to output a first generalization analysis report specifically includes:
[0079] Step S31: variable threshold generalization, that is, screening the strategy samples of preset performance according to the strategy threshold measurement and analysis report, and generalizing the optimal threshold of the strategy samples of preset performance according to the threshold generalization template; specifically: screening the variable with the third label of Y in the variable threshold measurement to perform variable threshold generalization, when performing variable threshold generalization, the analysis indicators mainly include: generalization time period, total sample size, front touch amount, strategy touch amount, repeated touch amount, additional touch amount, back touch amount, front touch rate, strategy touch rate, repeated touch rate, additional touch rate, back touch rate, overall overdue rate, additional touch overdue rate, back touch overall overdue rate, additional touch, additional touch risk multiple, additional touch rate dispersion, evaluation conclusion (whether it is recommended to go online), etc.;
[0080] Step S32: Automatically output variable threshold generalization analysis report (excel), that is, perform generalization analysis on the strategy samples that meet the preset conditions and their corresponding optimal thresholds according to the threshold generalization template to output the first generalization analysis report; the specific first generalization analysis report includes: generalization description, including strategy online screening criteria, threshold generalization indicator explanation, etc.; rule generalization, output strategy generalization related indicators for different time periods, such as strategy touch status, strategy effectiveness, evaluation conclusions, etc.; touch simulation analysis, analyze the strategy touch status in recent time (such as the past 10 weeks, the past 10 days).
[0081] In one embodiment, according to the first generalization analysis report, the strategy samples with expected performance are screened for online application, and multiple strategy samples with expected performance are merged, and generalization analysis is performed again to output the second generalization analysis report. Specifically, the steps are as follows: multiple strategy samples to be put online are merged into one strategy sample for generalization analysis. If at least two strategy samples with good effects are screened out for online use during variable threshold generalization, the two selected strategy samples are merged into a strategy sample with only two values of 0 and 1 (the value of touching the threshold is 1, otherwise the value is 0), and the strategy threshold is generalized using the threshold generalization template, and the second generalization analysis report is automatically output. Specifically, merging into one strategy sample is to reconstruct the variables. Each strategy sample can be understood as a single variable. The single variable is greater than the measured threshold and takes the value of 1, otherwise it takes the value of 0. In this way, the merged variable becomes a variable with only 0 and 1 values. Then, a value equal to 1 indicates that the strategy sample is touched, and a value equal to 0 indicates that the strategy sample is not touched. In this way, the performance and touch stability of the strategy sample can be simulated and analyzed. It is understandable that in order to analyze the overall performance of the strategies in this solution after they are launched, multiple strategy samples whose performances meet expectations are merged, and thus generalization analysis is performed again to output a second generalization analysis report.
[0082] It should be noted that the above threshold generalization template mainly includes the variable performance analysis and variable touch stability analysis of each time period, and will be automatically output in the form of a report. The calculation formula for the additional touch in the above embodiment is: (cum_bad / cum_good) / ((total_bad-cum_bad) / (total_good-cum_good))=(cumulative bad / cumulative good) / ((total bad-cumulative bad) / (total good-cumulative good));
[0083] Formula for calculating risk multiple: (number of bad samples in touched samples / number of touched samples) / (number of bad samples in full samples / full sample size); dispersion of additional touch rate: standard deviation of additional touch rate in different time periods / mean of additional touch rate in different time periods; front touch amount: touch amount of all existing rules; strategy touch amount: touch amount of rules to be launched; repeated touch amount: repeated touch amount of rules to be launched and existing rules; additional touch amount: amount of existing rules not touched by rules to be launched; back touch amount: total touch amount of all existing rules and rules to be launched; evaluation conclusion: if the additional touch risk multiples of the rules to be launched in different time periods are greater than a certain threshold (such as 3) and the dispersion of additional touch rate is less than or equal to a certain threshold (such as 0.4), it is recommended to go online, otherwise it is not recommended to go online.
[0084] The above are only embodiments of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the relevant field are aware of all the common technical knowledge in the technical field to which the invention belongs before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known systems should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for technicians in this field, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. A single-dimensional strategy automatic mining method, characterized in that: include: Obtain strategy sample data of different time spans required for single-dimensional strategy analysis; Perform threshold calculation on the policy sample data and output a policy threshold calculation analysis report, including the following contents: Performing descriptive statistical analysis on the strategy sample data and performing initial screening after the descriptive statistical analysis, and marking whether initial screening is performed as a first label, including: Select samples with mature target field performance in a certain period of time in the strategy sample data as training sets for threshold calculation, and automatically output the strategy threshold calculation analysis report; Based on the results of descriptive statistical analysis, the variables are initially screened, and the variables with missing rates below a certain threshold enter the next round of analysis, based on which whether the initial screening is performed is marked as the first label; The non-missing samples in the strategy sample data filtered by the first label are sorted from small to large according to the indicator values, and then subdivided and roughly divided, and the sample size and corresponding indicators corresponding to different bins are output; Among the samples in the preset range of non-missing sample bins, the threshold with the largest local indicator sample in the preset touch rate range is selected as the optimal threshold; Determine whether the local indicator samples and the global indicator samples meet the preset conditions, and mark the policy sample data that meet the preset conditions as the second label; according to reducing the correlation of the screening variable, mark the policy sample data that meet the preset conditions in the policy sample data marked as the second label as the third label; According to descriptive statistical analysis, optimal binning, screening of variables with good performance based on optimal binning, and threshold output strategy threshold calculation analysis report; According to the strategy threshold calculation analysis report, the strategy samples that meet the preset conditions and their corresponding optimal thresholds are screened for generalization analysis to output a first generalization analysis report; The analysis indicators include: generalization time period, total sample size, pre-touch quantity, strategic touch quantity, repeated touch quantity, additional touch quantity, post-touch quantity, pre-touch rate, strategic touch rate, repeated touch rate, additional touch rate, post-touch rate, overall overdue rate, additional touch overdue rate, post-touch overall overdue rate, additional touch, additional touch risk multiple, additional touch rate dispersion, and evaluation conclusions; The first generalization analysis report includes: generalization description, including strategy launch screening criteria and threshold generalization indicator explanation; rule generalization, output of strategy generalization related indicators in different time periods, including strategy touch-up situation, strategy effectiveness, and evaluation conclusions; touch-up simulation analysis, analysis of strategy touch-up situation in recent period; According to the first generalization analysis report, policy samples whose performance meets expectations are screened for online application, and multiple policy samples whose performance meets expectations are merged and generalization analysis is performed again to output a second generalization analysis report.
2. The single-dimensional strategy automatic mining method according to claim 1 is characterized in that: The strategy sample data includes independent variable data and dependent variable data.
3. The single-dimensional strategy automatic mining method according to claim 2 is characterized in that: The step of obtaining strategy sample data of different time spans required for single-dimensional strategy analysis includes: Obtaining an independent variable data dictionary; the independent variable data dictionary includes the English name of the strategy sample, the Chinese name of the strategy sample and the strategy sample type; According to the obtained independent variable data dictionary, the Chinese name of the strategy sample is matched for each independent variable data during strategy sample analysis.
4. The single-dimensional strategy automatic mining method according to claim 1 is characterized in that: The threshold calculation analysis report based on descriptive statistical analysis, optimal binning, screening of variables with good performance based on optimal binning, and threshold output strategy is specifically as follows: Based on descriptive statistical analysis, optimal binning, and screening of variables and thresholds with good performance based on optimal binning, the strategy threshold measurement and analysis report in Excel format is automatically output through Python.
5. The single-dimensional strategy automatic mining method according to claim 1 is characterized in that: The step of selecting strategy samples that meet preset conditions and their corresponding optimal thresholds for generalization analysis according to the threshold measurement and analysis report to output a first generalization analysis report includes: Filtering strategy samples with preset effectiveness according to the strategy threshold measurement and analysis report; A generalization analysis is performed on the optimal threshold of the strategy sample with the preset effectiveness screened out according to the threshold generalization template to output a first generalization analysis report.
6. The single-dimensional strategy automatic mining method according to claim 5 is characterized in that: The generalization analysis of the optimal threshold of the strategy sample with the preset effectiveness is performed according to the threshold generalization template as follows: Analyze and filter out the optimal thresholds of strategy samples with preset performance, touch rate and performance in different time periods, and automatically output indicators on whether to recommend strategy launch.
7. The single-dimensional strategy automatic mining method according to claim 5 is characterized in that: The steps of screening strategy samples whose performance meets expectations according to the first generalization analysis report for online application, merging multiple strategy samples whose performance meets expectations, and performing generalization analysis again to output a second generalization analysis report include: According to the first generalization analysis report, select strategy samples whose performance meets expectations and launch them online; Merge multiple strategy samples whose performance meets expectations into strategy sample data with touch threshold values of 0 and 1; Perform variable threshold generalization analysis according to the threshold generalization template and automatically output the second generalization analysis report.