Index fluctuation-based attribution analysis early warning method and system

Abstract

The invention relates to the technical field of risk control abnormity early warning, in particular to an attribution analysis early warning method and system based on index fluctuation. The server comprises a data real-time acquisition module, a data calculation module, a prediction module, an early warning module and a reason analysis module; the early warning module is used for calculating index fluctuation data of each dimension according to the data corresponding to each dimension in the multi-dimensional index real-time statistical information and the corresponding index predicted value, judging whether the corresponding index fluctuation data is greater than a corresponding preset fluctuation threshold value or not, if yes, judging that the data corresponding to the dimension is abnormal, and if not, judging that the data corresponding to the dimension is abnormal. Sending alarm information to a user, and storing data corresponding to the corresponding dimension into a database; and the reason analysis module is used for analyzing the data exception reason corresponding to the dimension according to the data corresponding to the corresponding dimension when the judgment result is that the data corresponding to the dimension is abnormal, generating a corresponding exception reason report and sending the exception reason report to the user.

Description

Technical Field

[0001] The present invention relates to the technical field of risk control anomaly warning, and in particular to an attribution analysis warning method and system based on indicator fluctuations. Background Art

[0002] In the operation of financial services, risk management departments bear the core responsibility of identifying and assessing potential risks for users and mitigating asset losses. Risk managers typically conduct comprehensive risk assessments before and during the loan process, using strategies such as operationalizing strategy models and implementing credit limit management.

[0003] However, in practice, frequent adjustments to policy models can easily lead to policy failures or other anomalies due to human error. For example, policy model adjustments can cause significant fluctuations in rejection rates, or even extreme cases where the approval rate suddenly spikes to 100% or drops to 0%. Furthermore, adjustments to limits such as credit limits and margin deposits can also trigger anomalies due to error. These anomalies are not only difficult to detect promptly, but once discovered, the lack of effective analytical tools makes it difficult to quickly pinpoint the specific causes.

[0004] Currently, risk control anomalies in risk control projects often rely on manual communication and investigation, which not only consumes a lot of time and costs, but also has low manual analysis efficiency, making it difficult to meet the efficient and accurate risk management needs of financial services. Summary of the Invention

[0005] One of the purposes of the present invention is to provide an attribution analysis and early warning method and system based on indicator fluctuations, which can achieve rapid response to risk control anomalies in risk control projects, thereby improving the efficiency of anomaly resolution.

[0006] In order to achieve the above objectives, an attribution analysis and early warning system based on indicator fluctuations is provided, including a server;

[0007] The server includes:

[0008] The real-time data collection module is used to send the corresponding decision details and data call details to Kafka after the risk control decision engine executes the decision or after the risk control variable calls the external data source;

[0009] The data calculation module is used to build the corresponding calculation window through Kafka's Kafka streams. Within the calculation window, it performs streaming aggregation on the decision detail data and data call detail data, and calculates the corresponding multi-dimensional real-time statistical information based on the streaming aggregated data.

[0010] The prediction module is used to calculate the indicator prediction value corresponding to each dimension in the multi-dimensional indicator real-time statistical information based on the calculated multi-dimensional indicator real-time statistical information and the preset indicator prediction value calculation formula;

[0011] The early warning module is used to calculate the indicator fluctuation data corresponding to each dimension based on the data corresponding to each dimension in the real-time statistical information of multi-dimensional indicators and the corresponding predicted values of each indicator, and determine whether the corresponding indicator fluctuation data is greater than the corresponding preset fluctuation threshold. If so, it is determined that the data corresponding to the dimension is abnormal, an alarm message is sent to the user, and the data corresponding to the corresponding dimension is stored in the database. Otherwise, it is determined that the data corresponding to the dimension is normal;

[0012] The cause analysis module is used to analyze the cause of the data anomaly corresponding to the dimension and generate a corresponding anomaly cause report based on the data corresponding to the corresponding dimension when the judgment result is that the data corresponding to the dimension is anomaly, and send it to the user.

[0013] The technical principles and effects of this solution: In this solution, after the risk control decision engine executes a decision or after a risk control variable calls an external data source, the real-time data collection module sends the corresponding decision details and data call details to Kafka. As a distributed message queue, Kafka efficiently receives and stores this data, providing a reliable data source for subsequent data processing. This step ensures that the system can obtain detailed information related to decisions and data calls in a timely manner, laying the foundation for subsequent analysis.

[0014] Kafka's Kafka streams are then used to construct a computation window. Within this computation window, streaming aggregation is performed on the decision-making and data call details retrieved from Kafka. Streaming aggregation is a technique for real-time processing and aggregation of continuous data streams. Aggregating data within a computation window enables more meaningful statistical information to be obtained. Based on this aggregated data, real-time statistics for corresponding multi-dimensional metrics are calculated. For example, statistical analysis can be performed based on multiple dimensions, such as time, user group, and business type, to generate real-time statistics for metrics such as click-through rate, conversion rate, and call frequency.

[0015] Based on the calculated real-time statistics of multi-dimensional indicators, the predicted indicator values corresponding to each dimension in the real-time statistics of multi-dimensional indicators are calculated based on preset indicator prediction value calculation formulas. These preset calculation formulas may be based on historical data, business rules, or statistical models. By analyzing and processing the current real-time statistical information, the possible indicator values of each dimension in the future are predicted.

[0016] Then, based on the data corresponding to each dimension in the real-time statistics of multi-dimensional indicators and the corresponding predicted values for each indicator, the indicator fluctuation data corresponding to each dimension is calculated. The indicator fluctuation data reflects the degree of difference between the actual data and the predicted data. The calculated indicator fluctuation data is compared with the corresponding preset fluctuation threshold. If the indicator fluctuation data exceeds the preset fluctuation threshold, the data corresponding to that dimension is judged to be abnormal, an alarm is sent to the user, and the data corresponding to the corresponding dimension is stored in the database for further analysis. Otherwise, the data corresponding to that dimension is judged to be normal.

[0017] When the early warning module identifies an anomaly in the data corresponding to a particular dimension, the cause analysis module analyzes the cause of the anomaly based on the corresponding data. This may involve in-depth data mining, comparative analysis with historical data, and streamlining of business processes. This analysis generates a corresponding anomaly cause report and sends it to the user, helping them understand the specific cause of the data anomaly so they can take appropriate action.

[0018] This solution leverages the real-time data collection module and Kafka's efficient data transmission capabilities to achieve millisecond-level monitoring of decision-making and data access processes. The data calculation module uses Kafka streams to build dynamic calculation windows for real-time streaming aggregation of data, synchronizing the update frequency of multi-dimensional metric statistics with the pace of business operations. The data calculation module supports cross-dimensional analysis across multiple dimensions, including time, region, user profiles, and business processes, breaking through the limitations of traditional single-dimensional monitoring.

[0019] Using technologies such as association rule mining and machine learning algorithms, we conduct in-depth tracing of abnormal data. When the early warning module triggers an alarm, the cause analysis module automatically retrieves historical records, business logs, and related indicators related to the abnormal data, and uses intelligent algorithms to quickly generate an abnormality cause report. This enables rapid response and resolution of risk anomalies within risk control projects.

[0020] Furthermore, the real-time statistical information of the multi-dimensional indicators includes the application approval rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution.

[0021] Beneficial effects: By calculating multi-dimensional indicator data, it is possible to comprehensively and accurately evaluate business conditions and system operations from different angles.

[0022] Furthermore, the preset indicator prediction calculation formula is:

[0023]

[0024] In the formula, y is the predicted value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment, a and b are the coefficients of linear regression, x is the actual value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment, and x is the actual value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment. i ,y i are the sample data already in the historical database, and are the predicted value and actual value of the i-th historical indicator corresponding to the historical database.

[0025] Beneficial effect: Using sample data in the historical database (x i ,y i ) to conduct linear regression analysis. By studying a large amount of historical data, we can capture the past patterns and trends of multi-dimensional indicators. For example, in the dimension of application approval rate, by combining the actual application approval rate values and corresponding predicted values in different time periods, we can calculate the coefficients a and b through linear regression, and thus make a relatively accurate prediction of the application approval rate at the current moment.

[0026] Substituting the actual value x of a particular dimension's indicator at the current moment into the formula to calculate the predicted value y, the system then compares it with the actual indicator value for that dimension. If the two differ significantly, meaning the indicator's fluctuation exceeds a preset fluctuation threshold, the system can issue a timely warning. For example, a significant deviation between the predicted and actual values of the interface call success rate could indicate a problem with the system's interaction with external interfaces, such as a network failure or interface update. Prompt warnings help technical personnel respond quickly and ensure stable system operation.

[0027] Furthermore, the server also includes a display module for drawing a corresponding line chart of the corresponding application pass rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution at the current moment when the result of the judgment is that the data corresponding to the dimension is abnormal, and visually displaying the result; each point in the line chart is associated with the corresponding abnormal reason report;

[0028] The drill-down module is used to receive the viewing feedback information corresponding to the user, and based on the viewing feedback information, determine the fluctuation point that the user needs to drill down in the line chart, and call up the abnormal reason report associated with the fluctuation time period centered on the fluctuation point, as well as the corresponding application pass rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution, and display them in a refined manner.

[0029] Beneficial Effects: When data anomalies are present, key indicators such as the application approval rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution are plotted and visualized as line charts. These line charts clearly demonstrate the changing trends of these indicators over time, allowing business personnel and decision makers to intuitively identify when and how these indicators experience unusual fluctuations. For example, by observing a line chart of the application approval rate, one can quickly identify the specific time point and magnitude of a drop in the approval rate. Compared to a simple data list, a visual approach is easier to understand and analyze, helping to quickly grasp the overall business situation.

[0030] Each point in the line chart is associated with a corresponding anomaly cause report. This design closely links abnormal fluctuations in indicators with the causes of the anomalies. When an anomaly is detected in a certain indicator, users can quickly understand the cause of the anomaly through this association. For example, if the success rate of an interface call success rate drops sharply at a certain point in time on the line chart, clicking that point will view the associated anomaly cause report. This could be due to an external interface server failure or network problem, allowing users to quickly locate the root cause of the problem and improve troubleshooting efficiency.

[0031] Receive user feedback and determine the fluctuation points in the line chart that require drilling down based on user needs. This allows users to conduct in-depth analysis of specific anomalies of interest, meeting the personalized analysis needs of different users. For example, a business manager may be more concerned about the abnormal fluctuations in the rule rejection rate within a specific time period. The drill-down function can provide an in-depth understanding of the detailed distribution of rejection reasons and changes in other related indicators within that time period, allowing for more accurate decision-making. Retrieve the abnormal cause report and corresponding indicator data associated with the fluctuation time period centered on the fluctuation point and present them in a refined manner. This refined presentation helps users fully and deeply understand the context in which the anomaly occurred and the comprehensive changes in related indicators.

[0032] The present invention also provides an attribution analysis and early warning method based on indicator fluctuations, using the above-mentioned attribution analysis and early warning system based on indicator fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a logic block diagram of an attribution analysis and early warning system based on indicator fluctuations in Example 1 of the present invention. DETAILED DESCRIPTION

[0034] The following is further described in detail through specific implementation methods:

[0035] Example 1

[0036] A method and system for early warning based on attribution analysis of indicator fluctuations, basically as follows Figure 1As shown, including the server;

[0037] The server includes:

[0038] The real-time data collection module is used to send the corresponding decision details and data call details to Kafka after the risk control decision engine executes the decision or after the risk control variable calls the external data source; in this embodiment, the decision details data includes products, scenarios, hit rules, and decision time data, and the data call details data includes interface provider, interface name, whether the call is successful, and call time data.

[0039] The data calculation module uses Kafka's Kafka streams to construct corresponding calculation windows. Within these calculation windows, it performs streaming aggregation on detailed decision and data call data. Based on this stream-aggregated data, it calculates real-time statistics for corresponding multi-dimensional indicators, including the approval rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution. For example, a 10-minute window contains multiple data entries. Assuming they are all related to the same product and scenario, with one decision being approved and one being rejected, at the end of the window, the calculation shows that there are two entries for this product and scenario, with a 50% approval rate.

[0040] In this embodiment, the corresponding detailed data is stream-aggregated, such as converting the risk control decision engine information into the following structure:

[0041] {"productNo":"9999888", / / Product No.

[0042] "sceneNo":"01", / / scene number

[0043] "finalDealType":"reject" / / Final decision result

[0044] "spendTime":241, / / Time spent

[0045] "creditAmount":500, / / Credit amount

[0046] "policySetId":"asdgsagash", / / Policy set ID

[0047] "Round":[

[0048] "roundName":"Round 1", / / Round name

[0049] "flag":"reject", / / round decision result

[0050] "policys":[

[0051] "policyId":"gashasfhd", / / policy id

[0052] "flag":"reject", / / strategy decision result

[0053] "hitRules":[

[0054] "ruleId":"hafdshadsfh", / / rule id

[0055] "flag": "reject" / / rule decision result]]]

[0056] "reason":[

[0057] "R1","R2" / / Rejection reason]}

[0058] Convert the data source call information into the following structure:

[0059] {"productNo":"9999888", / / Product No.

[0060] "sceneNo":"01", / / scene number

[0061] "groupId":"RH", / / Data source provider

[0062] "interfaceId":"CreditReport", / / Data source interface

[0063] "status": 0, / / Whether the call is successful

[0064] "executeTime":160, / / call duration}

[0065] The prediction module is used to calculate the indicator prediction value corresponding to each dimension in the multi-dimensional indicator real-time statistical information based on the calculated multi-dimensional indicator real-time statistical information and the preset indicator prediction value calculation formula;

[0066] The preset indicator prediction calculation formula is:

[0067]

[0068] In the formula, y is the predicted value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment, a and b are the coefficients of linear regression, x is the actual value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment, and x is the actual value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment. i ,yi are the sample data already in the historical database, and are the predicted value and actual value of the i-th historical indicator corresponding to the historical database.

[0069] The early warning module is used to calculate the indicator fluctuation data corresponding to each dimension based on the data corresponding to each dimension in the real-time statistical information of multi-dimensional indicators and the corresponding predicted values of each indicator, and to determine whether the corresponding indicator fluctuation data is greater than the corresponding preset fluctuation threshold. If so, it is determined that the data corresponding to the dimension is abnormal, an alarm message is sent to the user, and the data corresponding to the corresponding dimension is stored in the database. Otherwise, it is determined that the data corresponding to the dimension is not abnormal. In this embodiment, an alarm message is sent to the user. For example, if the rejection rate is 10% in the first 10 minutes and now suddenly becomes 90%, the system will issue a real-time alarm and can be reached through email, chat tools, telephone, etc.

[0070] The cause analysis module is used to analyze the cause of the data anomaly corresponding to the dimension and generate a corresponding anomaly cause report based on the data corresponding to the corresponding dimension when the result is that the data corresponding to the dimension is abnormal, and then send it to the user. When conducting an abnormality cause analysis, it is mainly divided into two types: dimensions and levels. The dimensions include manual ledgers and data fluctuations, and the causes are refined and divided into three basic categories: important, attention, and secondary. Manual operations are all important types. Data fluctuations are divided according to the type or proportion of fluctuations. For example, the time consumption of the data source usually has little effect on the results, so this type can be positioned as secondary. Others can be based on volatility, such as 10%-20% is secondary, 20%-35% is attention, and 35 or above is important. Risk managers can quickly locate according to different levels.

[0071] The server also includes a display module for drawing a corresponding line chart of the corresponding application pass rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution at the current moment when the result of the judgment is that the data corresponding to the dimension is abnormal, and visually displaying the corresponding line chart; each point in the line chart is associated with the corresponding abnormality reason report;

[0072] The drill-down module is used to receive the viewing feedback information corresponding to the user, and based on the viewing feedback information, determine the fluctuation point that the user needs to drill down in the line chart, and retrieve the abnormal reason report associated with the fluctuation time period centered on the fluctuation point, as well as the corresponding submission pass rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution, and perform a refined display. In this embodiment, drill down the data of the multi-dimensional fluctuation indicators of the time period to see which round rejection rate fluctuates more, which rule rejection rate fluctuates more, or which data source call success rate decreases more. You can also view the line chart of the corresponding dimension (for example, a certain rule rejection rate). Drill down the multi-dimensional operation records of the time period (quota limit adjustment records, engine component release records).

[0073] This embodiment also discloses an attribution analysis and early warning method based on indicator fluctuations, using the above-mentioned attribution analysis and early warning system based on indicator fluctuations.

[0074] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is excessively described here. Ordinary technicians in the relevant field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all 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 methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. 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. An attribution analysis and early warning system based on indicator fluctuations, characterized by: Including the server; The server includes: The real-time data collection module is used to send the corresponding decision details and data call details to Kafka after the risk control decision engine executes the decision or after the risk control variable calls the external data source; The data calculation module is used to build the corresponding calculation window through Kafka's Kafka streams. Within the calculation window, it performs streaming aggregation on the decision detail data and data call detail data, and calculates the corresponding multi-dimensional real-time statistical information based on the streaming aggregated data. The prediction module is used to calculate the indicator prediction value corresponding to each dimension in the multi-dimensional indicator real-time statistical information based on the calculated multi-dimensional indicator real-time statistical information and the preset indicator prediction value calculation formula; The early warning module is used to calculate the indicator fluctuation data corresponding to each dimension based on the data corresponding to each dimension in the real-time statistical information of multi-dimensional indicators and the corresponding predicted values of each indicator, and determine whether the corresponding indicator fluctuation data is greater than the corresponding preset fluctuation threshold. If so, it is determined that the data corresponding to the dimension is abnormal, an alarm message is sent to the user, and the data corresponding to the corresponding dimension is stored in the database. Otherwise, it is determined that the data corresponding to the dimension is normal; The cause analysis module is used to analyze the cause of the data anomaly corresponding to the dimension and generate a corresponding anomaly cause report based on the data corresponding to the corresponding dimension when the judgment result is that the data corresponding to the dimension is anomaly, and send it to the user.

2. The attribution analysis and early warning system based on indicator fluctuations according to claim 1 is characterized by: The real-time statistical information of the multi-dimensional indicators includes the application approval rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution.

3. The attribution analysis and early warning system based on indicator fluctuations according to claim 2 is characterized by: The preset indicator prediction calculation formula is: In the formula, y is the predicted value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment, a and b are the coefficients of linear regression, x is the actual value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment, and x is the actual value of the indicator corresponding to a certain dimension in the real-time statistical information of multi-dimensional indicators at the current moment. i ,y i are the sample data already in the historical database, and are the predicted value and actual value of the i-th historical indicator corresponding to the historical database.

4. The attribution analysis and early warning system based on indicator fluctuations according to claim 3 is characterized by: The server also includes a display module for drawing a corresponding line chart of the corresponding application pass rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution at the current moment when the result of the judgment is that the data corresponding to the dimension is abnormal, and visually displaying the corresponding line chart; each point in the line chart is associated with the corresponding abnormality reason report; The drill-down module is used to receive the viewing feedback information corresponding to the user, and based on the viewing feedback information, determine the fluctuation point that the user needs to drill down in the line chart, and call up the abnormal reason report associated with the fluctuation time period centered on the fluctuation point, as well as the corresponding application pass rate, rule rejection rate, round rejection rate, interface call success rate, and rejection reason distribution, and display them in a refined manner.

5. An attribution analysis and early warning method based on indicator fluctuations, characterized by: An attribution analysis and early warning system based on indicator fluctuations using any one of claims 1 to 4 above.

Images