Progressive cost optimization multi-objective strategy development method and system

Abstract

The invention relates to the technical field of financial risk control intelligent information, in particular to a multi-target strategy development method and system for progressive cost optimization, and the method comprises a data processing step which can carry out a first front refusal step and a second front refusal step in a serial or parallel manner. The system part at least comprises a project management module, a strategy workflow management module, a strategy result management module and an intelligent iteration module. According to the method, the cost optimization mode is gradually explored based on the risk control effect, and unified standard and process guidance are provided for the enterprise to evaluate the feature query priority; through millisecond-level data verification and second-level rule search, the breadth and depth of rule mining are improved, and meanwhile, the current strategy iteration efficiency is improved; the operation threshold of developers is reduced through the functions of low code configuration, top-speed copying and the like; the project type information management fits user habits and realizes strategy information structured management; the localized deployment system is helpful for precipitating expert experience and enabling sustainable development of mechanism strategy development.

Description

Technical Field

[0001] The present invention relates to the field of intelligent information technology for financial risk control, and in particular to a multi-objective strategy development method and system for progressive cost optimization. Background Art

[0002] In today's digital age, the financial industry's business models are constantly innovating, and transaction data is exploding. This has led to an increasing variety of risks faced by financial institutions, which are becoming increasingly complex and dynamic, placing extremely high demands on the accuracy, timeliness, and flexibility of risk control strategies.

[0003] For example, the widespread use of online payment has brought convenience, and criminals may use virtual accounts, false transactions and other means to engage in money laundering, card fraud and other activities, while traditional risk control strategies are difficult to identify and prevent these new risks in real time.

[0004] Traditional risk control strategy development relies primarily on manual data quality analysis and rule-finding by business personnel. Faced with massive amounts of financial data, manual processing capabilities are limited, consuming significant time and manpower. Furthermore, the scope and depth of rules that can be explored are limited, making it difficult to fully uncover potential risk characteristics and patterns, resulting in ineffective and inadequate strategies.

[0005] While the introduction of machine learning algorithms to assist in strategy development has improved efficiency and rule-mining capabilities to a certain extent, the complexity of the algorithmic models has made it difficult to effectively integrate expert business experience into the development process. Expert experience is crucial in risk control decision-making, but under traditional models, it cannot be fully integrated into the algorithmic models, thus limiting strategy optimization and iteration, making it difficult to achieve optimal risk control results.

[0006] Traditional strategy development often focuses on leveraging credit characteristics, with less attention paid to data query costs. In practice, large numbers of high-cost data queries can lead to diminishing returns on risk control investment. In some cases, the increased data query costs may not even lead to corresponding improvements in risk control effectiveness, making it difficult to achieve an optimal balance between risk control effectiveness and costs.

[0007] The existing related technologies have at least the following deficiencies:

[0008] 1. The dilemma of data-driven strategy development. While big data-based risk control strategy development boasts abundant data resources, it lacks effective methods to fully leverage this data. Traditional data analysis methods struggle to process large-scale, high-dimensional data, failing to quickly and accurately extract valuable insights from massive amounts of data. This results in a slow strategy development process and suboptimal results. For example, when assessing customer credit, traditional methods may fail to fully consider the various complex credit-related factors and their interrelationships, making it difficult to formulate accurate risk assessment strategies.

[0009] 2. Strategy iteration speed is insufficient to meet demand. The financial market environment is constantly changing, and risk profiles are constantly evolving. However, traditional risk control strategies have long development and update cycles, making them unable to respond promptly to market changes. This often leaves financial institutions in a passive position when facing new risk challenges, making them vulnerable to losses. For example, when new fraud patterns emerge or market fluctuations lead to changes in credit risk, traditional strategies may take a long time to adjust and optimize, failing to effectively mitigate risks in a timely manner.

[0010] To cope with the increasingly complex risk management environment, financial institutions urgently need intelligent, automated risk management strategy development technology. This technology should be able to automatically process and analyze large amounts of data, quickly generate effective risk management strategies, and be capable of real-time adjustment and optimization based on market changes and risk dynamics. While ensuring effective risk management, reducing costs has become a key concern for financial institutions. This requires more detailed consideration of cost factors during strategy development, optimizing data query and processing processes, and improving cost-effectiveness.

[0011] Existing financial risk control strategy development technology has many shortcomings when facing a complex and changing market environment and an ever-increasing amount of data. There is an urgent need for an innovative strategy development method and system that can progressively optimize risk costs and achieve a multi-objective balance, so as to enhance the risk control capabilities and competitiveness of financial institutions and ensure the stable and healthy development of the financial industry. Summary of the Invention

[0012] In view of this, the purpose of the present invention is to overcome the deficiencies of the prior art and provide a multi-objective strategy development method and system with progressive cost optimization, aiming to solve the problems of limited manual processing of data, difficulty in integrating with expert experience, and neglect of data query costs in the development of traditional risk control strategies. Through progressive cost optimization, invalid or high-risk traffic can be accurately identified and eliminated at different stages to achieve effective control of risk control costs. With the help of efficient traversal algorithms and various functions of intelligent systems, the efficiency of strategy iteration is improved, the breadth and depth of rule mining are expanded, and scientific guidance is provided for enterprises to evaluate feature query priorities. At the same time, "project-based" information management, low-code configuration and other means are used to adapt to user habits, lower the operating threshold, promote the deep integration of expert experience and machine computing power, and thus improve the accuracy and adaptability of risk control strategies, enhance the ability of financial institutions to cope with complex risk environments, ensure the stable and healthy development of financial services, and achieve the optimal balance between risk control effects and cost-effectiveness.

[0013] To achieve the above objectives, the present invention provides a first aspect of a multi-objective strategy development method for progressive cost optimization, comprising the following steps:

[0014] S1 data processing step, which is a prerequisite for the implementation of the remaining steps, is used to pre-process the original data to provide a data foundation suitable for strategy development. This includes setting the core variables in the data set for strategy development, cleaning the data, splitting the development set and validation set, performing exploratory statistical analysis on the data, and analyzing the original business indicators;

[0015] The S2 pre-rejection step is used to identify and eliminate extreme traffic under low-cost variables. By analyzing the statistical distribution of all low-cost or zero-cost variables and the risk performance of each quantile, the optimal extreme traffic rejection rule is found.

[0016] The S3 pre-rejection step is used to identify and eliminate traffic with poor actual risk performance among low-cost variables. By analyzing the statistical distribution of low-cost variables, one or more limiting parameters such as LIFT conditions, search step size, traversal termination condition threshold, and minimum sample size are set, and multiple rounds of strategy searches are performed until all rule sets that meet the requirements are found.

[0017] Steps S2 and S3 can be performed serially or in parallel.

[0018] Furthermore, the S1 data processing step specifically includes:

[0019] S101 Field Selection Step: Select fields from the dataset as candidate low-cost variables, target variables, current online decision results, challenged champion strategy results, and grouping variables required for strategy development; configure the allocation ratio of the development set and validation set in the overall sample set;

[0020] S102 Data analysis and cleaning step: Count the indicators of the selected variables in different time periods, and clean the data according to the indicator range and fluctuation, and the unique value type standard of specific variables;

[0021] S103 Risk Performance Calculation Steps: Calculate the number of samples, the number of passes, the pass rate, the overdue rate and performance rate of different target variables Yn within different time periods. If there are many months, historical statistics will be combined;

[0022] S104 Data Verification Step: Verify the cleaned data to check its integrity, accuracy, and consistency to ensure that the data meets the requirements of strategy development;

[0023] S105 Data conversion step: If the data is not in a uniform format or is not convenient for analysis, the data is converted into a different format or encoded so that it can be better used for subsequent statistical analysis.

[0024] Furthermore, the S2 pre-rejection step specifically includes:

[0025] S201 Extreme Traffic Analysis Steps: Calculate the values of P1, P2, P3, P4, P5, P25, P50, and P75 for the overall or grouped low-cost or zero-cost variables in each time period, and calculate the online decision-making success rate and the overdue rate of the main strategy target for samples within each quantile;

[0026] S202 Rejection rule setting step: Based on the current pass rate and bad debt rate requirements, set a limit of m% samples at the tail of any n low-cost variables, where n is greater than or equal to 1 and less than or equal to the number of low-cost variables, and m is based on risk requirements and current situation judgment;

[0027] S203 rule generation and selection step: Run rejection restrictions on the entire sample or each group sample separately to obtain multiple rules that meet the risk requirements, such as "reject if the tail 3% of any one variable among X1, X2, and X3 is hit". After selecting one or more rules, use "qz1_hit" to identify the overall hit situation of the selected rules, and use "challenge strategy result" to identify the comprehensive decision result of the previous rejection step 1;

[0028] S204 Rule Evaluation Step: Evaluate the generated rules. In addition to considering the rule's rejection effect on extreme traffic, the rule's impact on the overall sample distribution is analyzed to ensure the rationality and stability of the rules.

[0029] S205 rule adjustment step: Based on the result of the rule evaluation, the selected rule is adjusted, including fine-tuning the rejection condition or variable combination to optimize the performance of the rule.

[0030] Furthermore, the S3 pre-rejection step specifically includes:

[0031] S301 Distribution Observation Step: Calculate the P25, P50, and P75 values of the low-cost or zero-cost variable for the entire group or each group in each time period to observe the variable distribution;

[0032] S302 Dataset Applicability Judgment Step: Count the number of samples with performance in the development set, validation set, monthly development set, and monthly validation set, as well as the number of bad samples among the minimum samples with performance, to determine whether the dataset is suitable for developing rejection strategy rules based on the actual performance of the samples;

[0033] S303 parameter setting step: After confirming that this step is suitable for running, set different minimum LIFT requirements for the development set, validation set, monthly development set, and monthly validation set, and set the calculation step size and the minimum number of representative samples in each round of traversal results;

[0034] S304 rule traversal step: After arranging each low-cost variable in ascending order on the whole or each group sample, traverse multiple rules according to the calculation step length, and stop traversing after reaching the termination condition;

[0035] S305 Rule Screening Step: Screen all rules found in each sample set according to the development set, validation set, monthly development set, monthly validation set minimum LIFT requirements, and minimum performing sample requirements, and remove any non-compliant rules.

[0036] S306 rule sorting and selection steps:

[0037] S3061 Single variable rule sorting: For single variable rules, confirm the rules according to the LIFT priority or scale priority criteria;

[0038] S3062 Multivariate rule sorting: When comparing the rules of all variables, the final rule is selected according to the LIFT priority or scale priority criteria;

[0039] S307 Result Identification and Update Step: After selecting one or more rules based on the risk criteria, use "qz2_hit" to identify the overall hit status of all selected rules, and update the "Challenge Strategy Result" to the comprehensive decision result after the two pre-rejection steps;

[0040] S308 Strategy optimization step: Optimize the finalized rule set, consider the synergy between different rules, and improve the effectiveness and cost-effectiveness of the strategy.

[0041] Furthermore, in the S2 pre-rejection 1 step and the S3 pre-rejection 2 step, the analysis of the statistical distribution of low-cost or zero-cost variables includes calculating at least one statistic among the mean, median, mode, variance, standard deviation, skewness coefficient, and kurtosis coefficient of the variables, so as to more comprehensively understand the distribution characteristics of the variables; in the S3 pre-rejection 2 step, the cumulative rejection ratio in the termination condition can be dynamically adjusted according to the scale and risk requirements of the data set, and an upper limit value can be set for the traversal rounds to prevent excessive searching and waste of resources.

[0042] The second aspect of the present invention provides a multi-objective strategy development system for progressive cost optimization, which is used to implement the above method. The system at least includes: a project management module, a strategy workflow management module, a strategy result management module and an intelligent iteration module; wherein

[0043] The project management module is used to create new projects, filter, view basic information of created projects, view details and report information;

[0044] The strategic workflow management module is used to create new workflows, filter and view parameter information of created workflows; continue to edit unfinished workflows; and reuse part of the configuration of completed workflows to create new workflows;

[0045] The strategy result management module is used to view and download generated strategy result reports and corresponding data sets, and search and view historical strategy report records;

[0046] The intelligent iteration module is used to run pre-configured strategy development steps on the latest data at a fixed period to complete the cross-time period verification of strategy rules.

[0047] Furthermore, the project management module includes a new project function, a conditional screening function, a detail viewing function and a result viewing function:

[0048] The new project function: represents the process of analyzing a new data set until the first strategic workflow under the project is created;

[0049] The conditional filtering function supports multi-dimensional and multi-level filtering condition settings. It filters all created project records by project name, dataset name, creation time, project key indicator range, and associated business type, and displays the complete chain information of the policy workflow covered by the qualified projects, including the number of policy workflows, the number of policy results, the core data summary of the policy results, the creator, and detailed creation and update time information;

[0050] The View Details function provides in-depth project detail viewing capabilities, displaying comprehensive information about all created policy workflows under each project in a visual and structured manner, including the detailed logical flow of the selected policy development steps, real-time status monitoring data for each step, creator details, and creation and update time information;

[0051] The result viewing function: after at least one strategy workflow is completed and a report is generated, it comprehensively displays the detailed rule descriptions corresponding to all generated strategy results under the project, the basis and logical analysis of rule generation, detailed information of the submitter, and the pass rates of the original strategy and the new strategy in different time intervals, the overdue rate of multiple objective variables and the performance rate.

[0052] Furthermore, the strategic workflow management module includes the functions of creating a new workflow, filtering conditions, continuing development, viewing parameters, and copying:

[0053] The new workflow function: based on the existing data set under the project, according to the characteristics of the data set and historical strategy experience, a new strategy development workflow is created by adopting any step of the pre-rejection step 1 or the pre-rejection step 2;

[0054] The conditional filtering function: filters all created workflow records through multiple dimensional conditions such as workflow ID, update time, and workflow key parameter range, and views the selected strategy development steps, real-time status of each step, creator details, and creation and update time records of workflows that meet the conditions;

[0055] The continued development function provides intelligent recovery point positioning and historical operation backtracking for workflows in a temporary or calculation-completed state, allowing users to continue editing until they actively choose to end and generate a report.

[0056] The parameter viewing function displays the configuration parameters, update time, number of selected rules, and weight analysis information of the selected strategy development step for all created workflows in all states;

[0057] The copy function is used to select some or all development steps of a completed workflow, quickly copy its parameters and results, perform intelligent parameter adjustment and optimization based on the needs of the target workflow, and select other development steps to create a new strategic workflow.

[0058] Furthermore, the strategy result management module includes: condition screening function, report viewing function, data downloading function and strategy comparison function:

[0059] The conditional screening function: through the multi-dimensional conditions of workflow ID, policy result ID, update time, and policy key indicator range, all generated policy result records are screened to view the rule descriptions that meet the conditional results, the detailed execution process and results of the selected policy development steps, the creator's detailed information, and the records of creation and update time;

[0060] The report viewing function provides a variety of strategy result report viewing modes, including detailed text reports, visual data chart reports, and interactive dynamic reports. It comprehensively displays the overall report of the current strategy results, covering the detailed logic of the final rules of each selected strategy development step, and in-depth risk indicator analysis in the order and amount dimensions of the overall rule group;

[0061] The data download function supports users to download and view the final data set corresponding to the current result. Users can independently select the scope and format of the downloaded data, which includes the field information confirmed by the data processing process, the detailed hit status label and analysis data of the generated rules of each selected strategy development step, and the final challenge strategy result label corresponding to the current result;

[0062] The strategy comparison function is used to compare and analyze the multi-dimensional indicators of multiple strategy results, including pass rate, multi-target overdue rate, risk cost reduction rate, strategy risk performance stability index, and comprehensive improvement effect evaluation compared with online decision-making results.

[0063] Furthermore, the intelligent iteration module includes: creating new tasks, editing tasks, running control and viewing results functions:

[0064] The new task function is used to add intelligent iterative tasks, specifically including: automated offline data operations within the association organization, setting the task running frequency based on the scheduling rules and data update frequency of the data operation, selecting any steps of the pre-rejection 1 and pre-rejection 2 and configuring the corresponding parameters to realize system automation strategy development;

[0065] The task editing function provides a convenient and flexible task editing interface to modify and edit all parameters of the created task, including one or more of the task running frequency, associated data range, and strategy step parameters;

[0066] The operation control function: realizes the operation control of the created tasks, including opening or closing the created tasks; pausing, resuming the task operation and adjusting the task operation priority;

[0067] The result viewing function is used to view the strategy result report corresponding to each period, and at the same time, to view the fluctuation of the specified indicator within the specified period.

[0068] The present invention adopts the above technical solution, which has at least the following beneficial effects:

[0069] 1. The present invention adopts a progressive development method based on the gradual rejection of extreme traffic and actual performance, which can gradually explore cost optimization methods based on the risk control effect. In the process of strategy development, the extreme traffic under the low-cost variable is first identified and eliminated through the first step of pre-rejection, and then the traffic with poor actual risk performance is further screened out with the help of the second step of pre-rejection, so as to accurately remove those parts that may bring high risks and poor benefits, and realize effective optimization of costs from the perspective of risk control. For example, in the financial credit business scenario, it can avoid investing too much unnecessary audit costs in high-risk customer groups and reduce losses caused by potential bad debt risks, so that enterprises can optimize resource allocation to the greatest extent and reduce operating costs while ensuring that risks are controllable.

[0070] 2. The technical solution provided by the present invention uses a traversal algorithm to perform rule mining. Its advantage is that the algorithm traversal search depth is unlimited, which can expand the breadth and depth of rule mining. Combined with millisecond-level data verification and second-level rule search and risk indicator calculation, the overall speed of strategy development is greatly accelerated. This enables enterprises to quickly iterate and update strategies when facing market changes, business adjustments, etc., and quickly respond to business needs in different periods. For example, when new risk characteristics appear in the market or the business direction changes, enterprises can use the present invention to efficiently mine strategy rules that adapt to the new situation, improve the timeliness and adaptability of the strategy, maintain a dominant position in market competition, and far exceed the strategy iteration efficiency under existing technologies.

[0071] 3. The technical approach provided by the present invention can provide a unified standard and process guidance for enterprises to evaluate feature query priorities. In actual business operations, different personnel or departments often lack consistency and scientificity in their judgments on feature query priorities. The present invention, through clear development methods and steps, clearly defines how to analyze variable features, set rejection rules, and conduct multiple rounds of strategic searches. This provides a clear and unified reference for enterprises to evaluate which features are more worthy of priority queries and how to balance risks and costs. This avoids decision-making errors caused by differences in subjective judgments and helps improve the quality and efficiency of strategic decision-making for the entire enterprise.

[0072] 4. The intelligent system provided by this invention adopts a "project-based" information management approach. This management model is highly consistent with users' daily operations and thinking habits, allowing them to clearly and systematically categorize and manage various project-related strategic information. Furthermore, it enables structured management of strategic information, integrating scattered strategic data and process information across various links according to a specific logical framework. This facilitates user search, analysis, and subsequent utilization, effectively improving the value of information and management efficiency, offering significant advantages over traditional, loose information management models.

[0073] 5. The intelligent system provided by this invention has practical functions such as low-code configuration and extremely fast historical record replication. Low-code configuration means that developers can complete operations such as creating and adjusting policy workflows without writing a large amount of complex code, greatly simplifying the development process and shortening the development cycle. The extremely fast historical record replication function makes it easier for developers to reuse existing successful experiences or configurations, avoiding duplication of work. Even relatively inexperienced developers can quickly get started with policy development, effectively reducing the difficulty for developers to enter this field and improve the work efficiency and output quality of the overall development team.

[0074] 6. Developers of the intelligent system provided by this invention can more freely and flexibly adjust parameters based on actual business needs, data characteristics, and other factors, exploring a wider range of strategic possibilities. Furthermore, this approach effectively combines expert experience with machine computing power. Experts draw on their expertise and extensive experience to set strategic directions and key parameters, while machines leverage their powerful computing power to perform large-scale data processing, rule search, and calculations. The two complement each other, jointly empowering financial institutions' risk control efforts and creating a more scientific, precise, and adaptable risk control strategy system, effectively promoting the improvement of financial institutions' risk prevention and control capabilities and sustainable development.

[0075] The multi-objective strategy development method and supporting intelligent system for progressive cost optimization of the present invention demonstrate significant advantages in multiple key aspects, effectively solving many problems existing in the existing technology and providing strong support and guarantee for enterprises, especially financial institutions, in the fields of strategy development, risk control and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0077] Figure 1 It is a schematic flow chart of the multi-objective strategy development method of the present invention;

[0078] Figure 2 It is a module diagram of the multi-objective strategy development intelligent system of the present invention;

[0079] Figure 3 It is an entity relationship diagram of the multi-objective strategy development intelligent system of the present invention;

[0080] Figure 4 It is an execution flow chart of the multi-objective strategy development intelligent system of the present invention;

[0081] Figure 5 It is a schematic diagram of a new project of the multi-objective strategy development intelligent system of the present invention. DETAILED DESCRIPTION

[0082] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0083] Example 1

[0084] See also Figure 1 As shown, this embodiment provides a multi-objective strategy development method for progressive cost optimization, including the following steps:

[0085] S1 data processing step, which is a prerequisite for the implementation of the remaining steps, is used to pre-process the original data to provide a data foundation suitable for strategy development. This includes setting the core variables in the data set for strategy development, cleaning the data, splitting the development set and validation set, performing exploratory statistical analysis on the data, and analyzing the original business indicators;

[0086] The S2 pre-rejection step is used to identify and eliminate extreme traffic under low-cost variables. By analyzing the statistical distribution of all low-cost or zero-cost variables and the risk performance of each quantile, the optimal extreme traffic rejection rule is found.

[0087] The S3 pre-rejection step is used to identify and eliminate traffic with poor actual risk performance among low-cost variables. By analyzing the statistical distribution of low-cost variables, one or more limiting parameters such as LIFT conditions, search step size, traversal termination condition threshold, and minimum sample size are set, and multiple rounds of strategy searches are performed until all rule sets that meet the requirements are found.

[0088] Steps S2 and S3 can be performed serially or in parallel.

[0089] As an implementation manner, the S1 data processing step in this embodiment specifically includes:

[0090] S101 Field Selection Step: Select fields from the dataset as candidate low-cost variables X1, X2, X3, X4, ..., Xn, target variables Y1, Y2, Y3, ..., Yn, and the overdue amount variables corresponding to each target variable for strategy development; the current online decision results, the challenged champion strategy results, and the grouping variables; and configure the allocation ratio of the development set and the validation set in the overall sample set;

[0091] S102 Data analysis and cleaning steps: Count the count / missing / missingrate / unique / mean / min / max / quantiles of the selected variables in different time periods, and clean the data based on the indicator range and fluctuation, and the unique value type standard of specific variables;

[0092] S103 Risk Performance Calculation Steps: Calculate the number of samples, the number of passes, the pass rate, the overdue rate and performance rate of different target variables Yn within different time periods. If there are many months, historical statistics will be combined;

[0093] S104 Data Verification Step: Verify the cleaned data to check its integrity, accuracy, and consistency to ensure that the data meets the requirements of strategy development;

[0094] S105 Data conversion step: If the data is not in a uniform format or is not convenient for analysis, the data is converted into a different format or encoded so that it can be better used for subsequent statistical analysis.

[0095] As an implementation manner, the S2 pre-rejection step in this embodiment specifically includes:

[0096] S201 Extreme Traffic Analysis Steps: Calculate the values of P1, P2, P3, P4, P5, P25, P50, and P75 for the overall or grouped low-cost or zero-cost variables in each time period, and calculate the online decision-making success rate and the overdue rate of the main strategy target for samples within each quantile;

[0097] S202 Rejection rule setting step: Based on the current pass rate and bad debt rate requirements, set a limit of m% samples at the tail of any n low-cost variables, where n is greater than or equal to 1 and less than or equal to the number of low-cost variables, and m is based on risk requirements and current situation judgment;

[0098] S203 rule generation and selection step: Run rejection restrictions on the entire sample or each group sample separately to obtain multiple rules that meet the risk requirements, such as "reject if the tail 3% of any one variable among X1, X2, and X3 is hit". After selecting one or more rules, use "qz1_hit" to identify the overall hit situation of the selected rules, and use "challenge strategy result" to identify the comprehensive decision result of the previous rejection step 1;

[0099] S204 Rule Evaluation Step: Evaluate the generated rules. In addition to considering the rule's rejection effect on extreme traffic, the rule's impact on the overall sample distribution is analyzed to ensure the rationality and stability of the rules.

[0100] S205 rule adjustment step: Based on the result of the rule evaluation, the selected rule is adjusted, including fine-tuning the rejection condition or variable combination to optimize the performance of the rule.

[0101] As an implementation manner, the second step of S3 pre-rejection in this embodiment specifically includes:

[0102] S301 Distribution Observation Step: Calculate the P25, P50, and P75 values of the low-cost or zero-cost variable for the entire group or each group in each time period to observe the variable distribution;

[0103] S302 Dataset Applicability Judgment Step: Count the number of samples with performance in the development set, validation set, monthly development set, and monthly validation set, as well as the number of bad samples among the minimum number of samples with performance, to determine whether the dataset is suitable for developing a rejection strategy rule based on actual sample performance.

[0104] S303 parameter setting step: After confirming that this step is suitable for running, set different minimum LIFT requirements for the development set, validation set, monthly development set, and monthly validation set, and set the calculation step size and the minimum number of representative samples in each round of traversal results;

[0105] S304 rule traversal step: After arranging each low-cost variable in ascending order on the whole or each group sample, multiple rules are traversed according to the calculation step length, and the traversal is stopped after reaching the termination condition (cumulative rejection ratio or traversal rounds);

[0106] S305 Rule Screening Step: Screen all rules found in each sample set according to the development set, validation set, monthly development set, monthly validation set minimum LIFT requirements, and minimum performing sample requirements, and remove any non-compliant rules.

[0107] S306 rule sorting and selection steps:

[0108] S3061 Single-Variable Rule Sorting: For single-variable rules, prioritize the rules based on either LIFT or scale. For example, if two rules related to variable X1 are found, Rule 1 hits 100 people and has a LIFT of 4; Rule 2 hits 300 people and has a LIFT of 3.5. If LIFT is prioritized, Rule 1 with the highest LIFT should be selected; if scale is prioritized, Rule 2 with the largest scale should be selected.

[0109] S3062 Multivariate rule sorting: When comparing the rules of all variables, the final rule is selected according to the LIFT priority or scale priority criteria;

[0110] S307 Result Identification and Update Step: After selecting one or more rules based on the risk criteria, use "qz2_hit" to identify the overall hit status of all selected rules, and update the "Challenge Strategy Result" to the comprehensive decision result after the two pre-rejection steps;

[0111] S308 Strategy optimization step: Optimize the finalized rule set, consider the synergy between different rules, and improve the effectiveness and cost-effectiveness of the strategy.

[0112] In this embodiment, in the S2 pre-rejection 1 step and the S3 pre-rejection 2 step, the analysis of the statistical distribution of low-cost or zero-cost variables includes calculating at least one statistic among the mean, median, mode, variance, standard deviation, skewness coefficient, and kurtosis coefficient of the variables to more comprehensively understand the distribution characteristics of the variables; in the S3 pre-rejection 2 step, the cumulative rejection ratio in the termination condition can be dynamically adjusted according to the scale and risk requirements of the data set, and an upper limit value can be set for the traversal rounds to prevent excessive searching and waste of resources.

[0113] Example 2

[0114] like Figure 2 As shown, this embodiment also provides a multi-objective strategy development system for progressive cost optimization, which is used to implement the above method. The system at least includes: a project management module, a strategy workflow management module, a strategy result management module and an intelligent iteration module; wherein

[0115] The project management module is used to create new projects, filter, view basic information of created projects, view details and report information;

[0116] The strategic workflow management module is used to create new workflows, filter and view parameter information of created workflows; continue to edit unfinished workflows; and reuse part of the configuration of completed workflows to create new workflows;

[0117] The strategy result management module is used to view and download generated strategy result reports and corresponding data sets, and search and view historical strategy report records;

[0118] The intelligent iteration module is used to run pre-configured strategy development steps on the latest data at a fixed period to complete the cross-time period verification of strategy rules.

[0119] As a preferred implementation, the project management module in this embodiment includes a new project function, a conditional screening function, a detail viewing function, and a result viewing function:

[0120] The new project function: represents the process of analyzing a new data set until the first strategic workflow under the project is created;

[0121] The conditional filtering function supports multi-dimensional and multi-level filtering condition settings. It filters all created project records by project name, dataset name, creation time, project key indicator range, and associated business type, and displays the complete chain information of the policy workflow covered by the qualified projects, including the number of policy workflows, the number of policy results, the core data summary of the policy results, the creator, and detailed creation and update time information;

[0122] The View Details function provides in-depth project detail viewing capabilities, displaying comprehensive information about all created policy workflows under each project in a visual and structured manner, including the detailed logical flow of the selected policy development steps, real-time status monitoring data for each step, creator details, and creation and update time information;

[0123] The result viewing function: after at least one strategy workflow is completed and a report is generated, it comprehensively displays the detailed rule descriptions corresponding to all generated strategy results under the project, the basis and logical analysis of rule generation, detailed information of the submitter, and the risk control performance and business effects of the strategy results under the multi-dimensional evaluation system and the stability over different time periods.

[0124] As a preferred implementation, the policy workflow management module in this embodiment includes a new workflow function, a conditional screening function, a continued development function, a parameter viewing and a copy function:

[0125] The new workflow function: based on the existing data set under the project, according to the characteristics of the data set and historical strategy experience, a new strategy development workflow is created by adopting any step of the pre-rejection step 1 or the pre-rejection step 2;

[0126] The conditional filtering function: filters all created workflow records through multiple dimensional conditions such as workflow ID, update time, and workflow key parameter range, and views the selected strategy development steps, real-time status of each step, creator details, and creation and update time records of workflows that meet the conditions;

[0127] The continued development function provides intelligent recovery point positioning and historical operation backtracking for workflows in a temporary or calculation-completed state, allowing users to continue editing until they actively choose to end and generate a report.

[0128] The parameter viewing function displays the configuration parameters, update time, number of selected rules, and weight analysis information of the selected strategy development step for all created workflows in all states;

[0129] The copy function is used to select some or all development steps of a completed workflow, quickly copy its parameters and results, perform intelligent parameter adjustment and optimization based on the needs of the target workflow, and select other development steps to create a new strategic workflow.

[0130] As a preferred implementation, the policy result management module in this embodiment includes: condition screening function, report viewing function, data downloading function and policy comparison function:

[0131] The conditional screening function: through the multi-dimensional conditions of workflow ID, policy result ID, update time, and policy key indicator range, all generated policy result records are screened to view the rule descriptions that meet the conditional results, the detailed execution process and results of the selected policy development steps, the creator's detailed information, and the records of creation and update time;

[0132] The report viewing function provides a variety of strategy result report viewing modes, including detailed text reports, visual data chart reports, and interactive dynamic reports. It comprehensively displays the overall report of the current strategy results, covering the detailed logic of the final rules of each selected strategy development step, and in-depth risk indicator analysis in the order and amount dimensions of the overall rule group;

[0133] The data download function supports users to download and view the final data set corresponding to the current result. Users can independently select the scope and format of the downloaded data, which includes the field information confirmed by the data processing process, the detailed hit status label and analysis data of the generated rules of each selected strategy development step, and the final challenge strategy result label corresponding to the current result;

[0134] The strategy comparison function is used to compare and analyze the multi-dimensional indicators of multiple strategy results, including pass rate, multi-target overdue rate, risk cost reduction rate, strategy risk performance stability index, and comprehensive improvement effect evaluation compared with online decision-making results.

[0135] As a preferred embodiment, the intelligent iteration module in this embodiment includes: creating a new task, editing a task, running control and viewing results functions:

[0136] The new task function is used to add intelligent iterative tasks, specifically including: automated offline data operations within the association organization, setting the task running frequency based on the scheduling rules and data update frequency of the data operation, selecting any steps of the pre-rejection 1 and pre-rejection 2 and configuring the corresponding parameters to realize system automation strategy development;

[0137] The task editing function provides a convenient and flexible task editing interface to modify and edit all parameters of the created task, including one or more of the task running frequency, associated data range, and strategy step parameters;

[0138] The operation control function: realizes the operation control of the created tasks, including opening or closing the created tasks; pausing, resuming the task operation and adjusting the task operation priority;

[0139] The result viewing function is used to view the strategy result report corresponding to each period, and at the same time, to view the fluctuation of the specified indicator within the specified period.

[0140] In the above embodiment, the strategy development method includes data processing, pre-rejection one, and pre-rejection two, which are respectively used for cleaning and analyzing the original data, identifying low-cost extreme traffic, and identifying low-cost actual low-performance traffic. The strategy development intelligent system includes a project management module, a strategy workflow management module, a strategy result management, and an intelligent iteration module, which are respectively used for the integrated management of all strategy development workflows and results under a single data set, the management of all strategy development processes under a single project, the management of all strategy results under a single project, and automatic iteration task management based on a dynamic database. The present invention provides a development method and a supporting intelligent system that are suitable for multi-objective variable development scenarios and progressively optimize risk costs. It gradually explores cost optimization methods based on risk control effects, and provides a unified standard and process guidance for enterprises to evaluate feature query priorities; millisecond-level data verification and second-level rule search improve the breadth and depth of rule mining while improving the iteration efficiency of existing strategies; low-code configuration, ultra-fast copying and other functions reduce the operating threshold for developers; "project-based" information management fits user habits and realizes structured management of strategy information; localized deployment systems help to precipitate expert experience and enable the sustainable development of institutional strategy development.

[0141] Compared with the existing technology, the present invention provides a development method and a supporting intelligent system that gradually rejects and progressively optimizes risk costs based on extreme traffic and actual performance, gradually explores cost optimization methods based on risk control effects, and provides unified standards and process guidance for enterprises to evaluate feature query priorities; millisecond-level data verification, unlimited algorithm traversal search depth, second-level rule search and risk indicator calculation, which improve the breadth and depth of rule mining while improving the iteration efficiency of existing strategies; low-code configuration, ultra-fast copying and other functions reduce the operating threshold for developers; "project-based" information management conforms to user habits and realizes structured management of strategy information; localized deployment system helps to accumulate expert experience and enable the sustainable development of institutional strategy development.

[0142] In this embodiment, a multi-objective strategy development intelligent system with progressive cost optimization is designed to support the strategy development of the entire process in a "project" manner based on the implementation of the development method. Figure 3 The figure shows the entity relationships in the system, including Project 1 and Project 2. Each project corresponds to Dataset 1 and Dataset 2 respectively. Each dataset has multiple workflows (Workflow 1, Workflow 2...Workflow n), and each workflow corresponds to multiple policy results (Policy Result 1, Policy Result 2...Policy Result n).

[0143] The system includes four entities: projects, data sets, policy workflows, and policy results; among them, projects are one-to-one with data sets, one-to-many with policy workflows and policy results, and one-to-many with policy workflows and policy results. Figure 3 The corresponding relationship between entities is presented intuitively. For example, Project 1 corresponds to Dataset 1, Dataset 1 corresponds to multiple workflows such as Workflow 1 and Workflow 2, and Workflow 1 corresponds to multiple policy results such as Policy Result 1 and Policy Result 2.

[0144] Attachment Figure 4 Three application scenarios of the intelligent system are demonstrated, namely Scenario 1: Creating a New Project, Scenario 2: Managing Unfinished Projects, and Scenario 3: Managing Completed Projects, and the corresponding operation procedures are marked under each scenario.

[0145] Scenario 1: Create a new project. This includes the data processing process described in the claims and the first workflow sub-process under the new project. The data processing process is used to import the data set corresponding to the project, set fields, and analyze data characteristics and business performance. The new workflow sub-process refers to the strategy development process that selects any step between pre-rejection 1 and pre-rejection 2 described in the claims. During the development and configuration process, it is allowed to temporarily save and exit and then continue development;

[0146] Scenario 2: Managing unfinished projects. When a project has no policy results and the project status is "unfinished," you can use the "View Details" function to enter the policy workflow management module under that project to view the corresponding workflow records, continue developing unfinished work, or create a new workflow.

[0147] Scenario 3: Manage Completed Projects. When a project has at least one policy result and the project status is "Complete," click "View Results" to access the policy result management module for that project to view the corresponding policy result records, view the corresponding report, or download the corresponding dataset.

[0148] Example 3

[0149] like Figure 5As shown, this embodiment describes in detail the process of creating a new project, covering the new workflow process of S1 data processing and any steps in S2 pre-rejection 1 and S3 pre-rejection 2.

[0150] 1. S1: Data Processing

[0151] This step is divided into four sub-steps: basic information, verification, EDA, and business interpretation.

[0152] (1) Basic information

[0153] Project and dataset naming

[0154] First, you need to enter the project and dataset names. This helps to clearly identify and manage the project. For example, the project name could be "Credit Risk Control Strategy Optimization Project" and the dataset name could be "2024 Q3 Credit Data."

[0155] Data import and variable selection

[0156] After importing the data file, select variables from the dataset that are meaningful for strategy development. These variables will serve as the basis for subsequent analysis and strategy development.

[0157] Development set and validation set splitting and data removal rule setting

[0158] Based on business needs and data characteristics, appropriately set the development and validation set split ratio. For example, 70% of the data could be allocated to the development set and 30% to the validation set. Also, set data exclusion rules to ensure the quality of data entering subsequent analysis, such as removing data that is obviously abnormal or inconsistent with business logic.

[0159] (2) Verification

[0160] The system background automatically performs data verification to ensure that the set variables conform to the business logic.

[0161] Target variable enumeration value verification

[0162] Check whether the enumeration value of the target variable meets the preset requirements. For example, in a credit risk control scenario, the enumeration value of the target variable usually only has 0 (representing not overdue), 1 (representing overdue), null (representing missing data), etc.

[0163] Sample primary key uniqueness check

[0164] Ensure the uniqueness of the sample primary key to avoid data duplication that may interfere with subsequent analysis. For example, each customer's identity identifier (such as their ID number) should be unique within the dataset.

[0165] (3) EDA (Exploratory Data Analysis)

[0166] Calculation of statistical distribution indicators

[0167] Calculates a series of statistical distribution indicators for the selected variable, including count, missing, missing rate, unique, mean, min, max, and pn. For example, for the customer age variable, calculate its mean as 35, minimum as 20, and maximum as 60.

[0168] Statistics on fluctuations of important indicators

[0169] Statistics on fluctuations in important indicators such as missing and missing rate over the past four months are collected. This helps us understand the stability and changing trends of the data. If the missing rate of a variable is found to have gradually increased over the past four months, further analysis may be needed to determine the cause.

[0170] (IV) Business Interpretation

[0171] Raw data business indicator statistics

[0172] Calculate the approval rate and overdue rate of different target variables within each time period of the original data. For example, calculate the approval rate and overdue rate of credit applications on a monthly basis.

[0173] Combined statistics of multi-period business indicators

[0174] Combine the business indicators of each time period into 4 months and then make comprehensive statistical business indicators to gain a more comprehensive understanding of long-term business performance.

[0175] 2. S2: Pre-rejection 1

[0176] Users can choose to skip this step and proceed to S3 Pre-Rejection 2. If they do not skip this step, they need to configure the parameters of this step. The calculation parameters and steps for the overall and group development scenarios are the same, including setting variables, viewing EDA results, configuring rejection methods, and outputting results.

[0177] (1) Setting variables

[0178] Low-cost variables and main target variable settings

[0179] Based on the variables selected in S1 Data Processing - Basic Information, further determine the low-cost variables and primary target variables required for this step. For example, in credit risk control, the customer's age, occupation, etc. may be used as low-cost variables, and overdue repayment status as the primary target variable.

[0180] Grouping variable selection (if group development)

[0181] If you are developing a grouped strategy, select the grouping variable for this step based on the grouping variable selected in the data processing step. For example, when developing a risk control strategy by region, select the region variable as the grouping variable.

[0182] (2) View EDA results

[0183] The system will collect relevant data for each low-cost variable.

[0184] Key numerical statistics

[0185] Count the missing rate, P1, P2, P5, and other values of each low-cost variable. For example, the missing rate of a low-cost variable is 5%, and the P1 value is 10.

[0186] Quantile sample index statistics

[0187] Calculate the pass rate and overdue rate of the samples at the corresponding quantiles based on the main target. For example, at the P2 quantile of the low-cost variable, the pass rate of the samples is 70% and the overdue rate is 10%.

[0188] (3) Rejection method configuration

[0189] Rejection rule determination

[0190] Set "Hit the tail n% of any m low-cost variables" as a rejection rule, and the number of configured rules should not exceed the number of low-cost variables. For example, set "Hit the tail 10% of any two low-cost variables" as a rejection rule.

[0191] (IV) Output results

[0192] Rule Generation

[0193] The system automatically runs the pre-rejection 1 in the strategy development method and generates multiple rules that meet the configuration.

[0194] Rule selection and effect review

[0195] Users can select any n rules and click "View Overall Effect of Selected Rules." The system will then calculate metrics like the pass rate and overdue rate based on the combined effect of all rules on the dataset. For example, if three rules are selected, the system will calculate a 60% pass rate and an 8% overdue rate for the dataset.

[0196] Dataset labeling

[0197] The label "qz1_hit" is added to the dataset to indicate the hit status of the rules selected in this stage, and the label "challenge strategy result" indicates the final decision result of this stage.

[0198] Next step selection

[0199] Save operation: If the user chooses to save, the current workflow status changes to "Save". You can then click "Continue Development" in the policy workflow management module to continue this process.

[0200] Enter the Pre-Rejection 2 operation: If the user chooses to enter the Pre-Rejection 2 operation, the decision result of the rule selected in this step will be applied and the S3 Pre-Rejection 2 step will be continued.

[0201] End and generate report: If you select End and generate report, the workflow status changes to Completed, the selected module includes this step, and a policy result record corresponding to the selected rule is generated. You can then view the report and download the latest dataset in the Policy Result Management module.

[0202] S3: Pre-rejection 2

[0203] The calculation parameters and steps for the overall and group development scenarios are the same, and also include four steps: setting variables, viewing EDA results, configuring rejection methods, and outputting results.

[0204] (1) Setting variables

[0205] Low-cost variables and main target variable settings

[0206] Similar to the variable setting operation in S2 pre-rejection 1, based on the variables selected in S1 data processing-basic information, the low-cost variables and main target variables required for this step are further determined.

[0207] Grouping variable selection (if group development)

[0208] If grouping development is performed, select the grouping variable for this step based on the grouping variable selected in the data processing step.

[0209] (2) View EDA results

[0210] The system will collect relevant data for each low-cost variable.

[0211] Key numerical statistics

[0212] Calculate the missing rate, P25, P50, P75, and other values for each low-cost variable. For example, the missing rate of a low-cost variable is 3%, the P25 value is 20, the P50 value is 30, and the P75 value is 40.

[0213] Quantile sample index statistics

[0214] Calculate the pass rate and overdue rate of the samples at the corresponding quantiles based on the main target. For example, at the P25 quantile of the low-cost variable, the pass rate is 75% and the overdue rate is 7%.

[0215] (3) Reject parameter configuration

[0216] Sample verification requirements configuration

[0217] Configure sample verification requirements to verify whether the sample meets the requirements before traversal calculation. If the sample does not meet the requirements, the calculation will fail directly; if it meets the requirements, the traversal calculation and strategy selection will be performed according to the strategy selection parameters.

[0218] Strategy selection requires parameter configuration

[0219] Configure the strategy selection requirement parameters to determine how to select the appropriate strategy from the traversal calculation results.

[0220] (IV) Output results

[0221] Rule Generation

[0222] The system automatically runs the pre-rejection 2 in the strategy development method and generates multiple rules that meet the configuration.

[0223] Rule selection and effect review

[0224] Users can select any n rules and click "View the overall effect of selected rules". The system will calculate indicators such as the pass rate and overdue rate based on the combined effect of all rules on the data set.

[0225] Adding and updating dataset labels

[0226] Add the label "qz2_hit" to the dataset to indicate the hit status of the rules selected in this stage, and update the "Challenge Strategy Result" label to indicate the final decision result of this stage.

[0227] Next step selection

[0228] Save operation: If the user chooses to save, the current workflow status changes to "Save". You can then click "Continue Development" in the policy workflow management module to continue this process.

[0229] End and generate report: If you select End and generate report, the workflow status changes to Completed, the selected module includes this step, and a policy result record corresponding to the selected rule is generated. You can then view the report and download the latest dataset in the Policy Result Management module.

[0230] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A multi-objective strategy development method for progressive cost optimization, characterized by: The following steps are involved: S1 data processing step, which is a prerequisite for the implementation of the remaining steps, is used to pre-process the original data to provide a data foundation suitable for strategy development. This includes setting the core variables in the data set for strategy development, cleaning the data, splitting the development set and validation set, performing exploratory statistical analysis on the data, and analyzing the original business indicators; The S2 pre-rejection step is used to identify and eliminate extreme traffic under low-cost variables. By analyzing the statistical distribution of all low-cost or zero-cost variables and the risk performance of each quantile, the optimal extreme traffic rejection rule is found. The S3 pre-rejection step is used to identify and eliminate traffic with poor actual risk performance among low-cost variables. By analyzing the statistical distribution of low-cost variables, one or more limiting parameters such as LIFT conditions, search step size, traversal termination condition threshold, and minimum sample size are set, and multiple rounds of strategy searches are performed until all rule sets that meet the requirements are found. Steps S2 and S3 can be performed serially or in parallel.

2. The method according to claim 1, wherein: The S1 data processing step specifically includes: S101 Field Selection Step: Select fields from the dataset as candidate low-cost variables, target variables, current online decision results, challenged champion strategy results, and grouping variables required for strategy development; configure the allocation ratio of the development set and validation set in the overall sample set; S102 Data analysis and cleaning step: Count the indicators of the selected variables in different time periods, and clean the data according to the indicator range and fluctuation, and the unique value type standard of specific variables; S103 Risk Performance Calculation Steps: Calculate the number of samples, the number of passes, the pass rate, the overdue rate and performance rate of different target variables Yn within different time periods. If there are many months, historical statistics will be combined; S104 Data Verification Step: Verify the cleaned data to check its integrity, accuracy, and consistency to ensure that the data meets the requirements of strategy development; S105 Data conversion step: If the data is not in a uniform format or is not convenient for analysis, the data is converted into a different format or encoded so that it can be better used for subsequent statistical analysis.

3. The method according to claim 1, wherein: The S2 pre-rejection step specifically includes: S201 Extreme Traffic Analysis Steps: Calculate the values of P1, P2, P3, P4, P5, P25, P50, and P75 for the overall or grouped low-cost or zero-cost variables in each time period, and calculate the online decision-making success rate and the overdue rate of the main strategy target for samples within each quantile; S202 Rejection rule setting step: Based on the current pass rate and bad debt rate requirements, set a limit of m% samples at the tail of any n low-cost variables, where n is greater than or equal to 1 and less than or equal to the number of low-cost variables, and m is based on risk requirements and current situation judgment; S203 Rule Generation and Selection Step: Run rejection restrictions on the entire sample or each grouped sample to obtain multiple rules that meet the risk requirements, such as "Reject if the tail 3% of any variable among X1, X2, and X3 is hit." After selecting one or more rules, use "qz1_hit" to identify the overall hit status of the selected rules, and use "Challenge Strategy Result" to identify the comprehensive decision result of the previous rejection step 1; S204 Rule Evaluation Step: Evaluate the generated rules. In addition to considering the rule's rejection effect on extreme traffic, the rule's impact on the overall sample distribution is analyzed to ensure the rationality and stability of the rules. S205 rule adjustment step: Based on the result of the rule evaluation, the selected rule is adjusted, including fine-tuning the rejection condition or variable combination to optimize the performance of the rule.

4. The method according to claim 1, wherein The S3 pre-rejection step specifically includes: S301 Distribution Observation Step: Calculate the P25, P50, and P75 values of the low-cost or zero-cost variable for the entire group or each group in each time period to observe the variable distribution; S302 Dataset Applicability Judgment Step: Count the number of samples with performance in the development set, validation set, monthly development set, and monthly validation set, as well as the number of bad samples among the minimum samples with performance, to determine whether the dataset is suitable for developing rejection strategy rules based on the actual performance of the samples; S303 parameter setting step: After confirming that this step is suitable for running, set different minimum LIFT requirements for the development set, validation set, monthly development set, and monthly validation set, and set the calculation step size and the minimum number of representative samples in each round of traversal results; S304 rule traversal step: After arranging each low-cost variable in ascending order on the whole or each group sample, traverse multiple rules according to the calculation step length, and stop traversing after reaching the termination condition; S305 Rule Screening Step: Screen all rules found in each sample set according to the development set, validation set, monthly development set, monthly validation set minimum LIFT requirements, and minimum performing sample requirements, and remove any non-compliant rules. S306 rule sorting and selection steps: S3061 Single variable rule sorting: For single variable rules, confirm the rules according to the LIFT priority or scale priority criteria; S3062 Multivariate rule sorting: When comparing the rules of all variables, the final rule is selected according to the LIFT priority or scale priority criteria; S307 Result Identification and Update Step: After selecting one or more rules based on the risk criteria, use "qz2_hit" to identify the overall hit status of all selected rules, and update the "Challenge Strategy Result" to the comprehensive decision result after the two pre-rejection steps; S308 Strategy optimization step: Optimize the finalized rule set, consider the synergy between different rules, and improve the effectiveness and cost-effectiveness of the strategy.

5. The method according to claim 1, wherein In the S2 pre-rejection 1 step and the S3 pre-rejection 2 step, the analysis of the statistical distribution of low-cost or zero-cost variables includes calculating at least one statistic among the mean, median, mode, variance, standard deviation, skewness coefficient, and kurtosis coefficient of the variables to more comprehensively understand the distribution characteristics of the variables; in the S3 pre-rejection 2 step, the cumulative rejection ratio in the termination condition can be dynamically adjusted according to the scale and risk requirements of the data set, and an upper limit value can be set for the traversal rounds to prevent excessive searching and waste of resources.

6. A multi-objective strategy development system for progressive cost optimization, characterized by: The system is used to implement the method described in any one of claims 1 to 5 above, and the system comprises at least: a project management module, a strategy workflow management module, a strategy result management module and an intelligent iteration module; wherein The project management module is used to create new projects, filter, view basic information of created projects, view details and report information; The strategic workflow management module is used to create new workflows, filter and view parameter information of created workflows; continue to edit unfinished workflows; and reuse part of the configuration of completed workflows to create new workflows; The strategy result management module is used to view and download generated strategy result reports and corresponding data sets, and search and view historical strategy report records; The intelligent iteration module is used to run pre-configured strategy development steps on the latest data at a fixed period to complete the cross-time period verification of strategy rules.

7. The system according to claim 6, characterized in that: The project management module includes the functions of creating a new project, filtering conditions, viewing details and viewing results: The new project function: represents the process of analyzing a new data set until the first strategic workflow under the project is created; The conditional filtering function supports multi-dimensional and multi-level filtering condition settings. It filters all created project records by project name, dataset name, creation time, project key indicator range, and associated business type, and displays the complete chain information of the policy workflow covered by the qualified projects, including the number of policy workflows, the number of policy results, the core data summary of the policy results, the creator, and detailed creation and update time information; The View Details function provides in-depth project detail viewing capabilities, displaying comprehensive information about all created policy workflows under each project in a visual and structured manner, including the detailed logical flow of the selected policy development steps, real-time status monitoring data for each step, creator details, and creation and update time information; The result viewing function: after at least one strategy workflow is completed and a report is generated, it comprehensively displays the detailed rule descriptions corresponding to all generated strategy results under the project, the basis and logical analysis of rule generation, detailed information of the submitter, and the risk control performance and business effects of the strategy results under the multi-dimensional evaluation system and their stability over different time periods.

8. The system according to claim 6, characterized in that: The strategic workflow management module includes the functions of creating a new workflow, filtering conditions, continuing development, viewing parameters, and copying: The new workflow function: based on the existing data set under the project, according to the characteristics of the data set and historical strategy experience, a new strategy development workflow is created by adopting any step of the pre-rejection step 1 or the pre-rejection step 2; The conditional filtering function: filters all created workflow records through multiple dimensional conditions such as workflow ID, update time, and workflow key parameter range, and views the selected strategy development steps, real-time status of each step, creator details, and creation and update time records of workflows that meet the conditions; The continued development function provides intelligent recovery point positioning and historical operation backtracking for workflows in a temporary or calculation-completed state, allowing users to continue editing until they actively choose to end and generate a report. The parameter viewing function displays the configuration parameters, update time, number of selected rules, and weight analysis information of the selected strategy development step for all created workflows in all states; The copy function is used to select some or all development steps of a completed workflow, quickly copy its parameters and results, perform intelligent parameter adjustment and optimization based on the needs of the target workflow, and select other development steps to create a new strategic workflow.

9. The system according to claim 6, characterized in that: The strategy result management module includes: condition screening function, report viewing function, data downloading function and strategy comparison function: The conditional screening function: through the multi-dimensional conditions of workflow ID, policy result ID, update time, and policy key indicator range, all generated policy result records are screened to view the rule descriptions that meet the conditional results, the detailed execution process and results of the selected policy development steps, the creator's detailed information, and the records of creation and update time; The report viewing function provides a variety of strategy result report viewing modes, including detailed text reports, visual data chart reports, and interactive dynamic reports. It comprehensively displays the overall report of the current strategy results, covering the detailed logic of the final rules of each selected strategy development step, and in-depth risk indicator analysis in the order and amount dimensions of the overall rule group; The data download function supports users to download and view the final data set corresponding to the current result. Users can independently select the scope and format of the downloaded data, which includes the field information confirmed by the data processing process, the detailed hit status label and analysis data of the generated rules of each selected strategy development step, and the final challenge strategy result label corresponding to the current result; The strategy comparison function is used to compare and analyze the multi-dimensional indicators of multiple strategy results, including pass rate, multi-target overdue rate, risk cost reduction rate, strategy risk performance stability index, and comprehensive improvement effect evaluation compared with online decision-making results.

10. The system according to claim 6, characterized in that: The intelligent iteration module includes: creating a new task function, editing a task function, running control function and viewing result function: The new task function is used to add intelligent iterative tasks, specifically including: automated offline data operations within the association organization, setting the task running frequency based on the scheduling rules and data update frequency of the data operation, selecting any steps of the pre-rejection 1 and pre-rejection 2 and configuring the corresponding parameters to realize system automation strategy development; The task editing function provides a convenient and flexible task editing interface to modify and edit all parameters of the created task, including one or more of the task running frequency, associated data range, and strategy step parameters; The operation control function: realizes the operation control of the created tasks, including opening or closing the created tasks; pausing, resuming the task operation and adjusting the task operation priority; The result viewing function is used to view the strategy result report corresponding to each period, and at the same time, to view the fluctuation of the specified indicator within the specified period.

Images