Insurance business evaluation system based on big data analysis

Abstract

The invention relates to the technical field of insurance service evaluation systems, and particularly discloses an insurance service evaluation system based on big data analysis, which comprises a data acquisition module, a risk evaluation module, an insurance premium calculation module and a claim settlement evaluation module, the risk assessment module is configured to analyze the customer risk condition based on the insurance service data set: firstly, calculating a reference index reflecting the historical risk of the customer, and then obtaining a risk coefficient through weighted fusion and dimensionless processing in combination with the risk level of the customer group. According to the invention, a complete customer risk portrait is constructed by integrating internal and external multi-source data, dynamic risk assessment and personalized insurance premium adjustment are realized, claim settlement assessment efficiency and fraud identification capability are improved by means of a machine learning model, and each module has dynamic update and expansion capability. The problems of low precision, poor adaptability, low efficiency and the like of a traditional evaluation mode are effectively solved, accurate risk control of insurance companies is assisted, the customer experience is optimized, and the operation cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of insurance business evaluation system technology, specifically an insurance business evaluation system based on big data analysis. Background Technology

[0002] In the process of digital transformation in the insurance industry, traditional insurance business assessment models have gradually revealed many limitations, making it difficult to meet the industry's needs for accurate risk control, personalized services, and efficient claims settlement. The main problems are reflected in the following three aspects:

[0003] 1. Traditional insurance business assessments often rely heavily on internal policy data and basic customer information, such as static data like policy type, coverage amount, and age, neglecting the impact of external environmental factors (such as the frequency of regional weather disasters and geographical accident rates) and dynamic customer behavior (such as payment timeliness and long-term policy loyalty). For example, in auto insurance assessments, risk is determined solely based on vehicle model and customer age, without considering the impact of the annual frequency of heavy rains in the insured area on the accident rate; in life insurance assessments, the willingness to fulfill obligations reflected in the customer's long-term payment records is not considered, resulting in incomplete risk profiles and the potential for misjudging high-risk customers as low-risk or overpricing low-risk customers.

[0004] 2. Traditional assessments often rely on manual rule-based judgments or simple statistical models, such as classifying risk levels based on fixed thresholds. These methods struggle to address the complex non-linear risk relationships and dynamically changing market environments inherent in insurance business. For instance, economic fluctuations (such as changes in GDP growth rates) can affect customers' ability to pay, thereby altering premium default risk. However, traditional models do not incorporate such dynamic factors into their assessments. Furthermore, manual rule updates are often outdated, failing to respond promptly to new risk patterns (such as new forms of insurance fraud), resulting in poor timeliness of risk warnings and a lack of data support for premium adjustments and claims decisions.

[0005] 3. Traditional claims processes rely on manual review of claim materials, which results in long review periods and significant human error. This is especially problematic in a large volume of claims, making it difficult to quickly identify fraud risks (such as false accident claims or duplicate claims). Furthermore, claims assessments are not linked to customers' historical risks and dynamic behaviors, making it impossible to accurately quantify the level of claim risk. This can easily lead to delays in processing legitimate cases or omissions in handling fraudulent cases, impacting customer experience and increasing insurance company payout costs. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an insurance business evaluation system based on big data analysis, which solves the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an insurance business evaluation system based on big data analysis, comprising a data acquisition module, a risk assessment module, a premium calculation module, and a claims assessment module;

[0008] The data acquisition module is configured to collect insurance-related multi-source data, and construct a structured insurance business dataset through data cleaning, standardization and feature extraction processing to provide a high-quality data foundation for subsequent evaluation;

[0009] The risk assessment module is configured to analyze customer risk profiles based on insurance business datasets: first, it calculates benchmark indicators reflecting a customer's historical risk; then, it combines these with the risk levels of the customer group, and obtains risk coefficients through weighted fusion and dimensionless processing. Preset risk threshold ,Will and If compared, Exceed If a high-risk hazard is identified, an early warning instruction is sent to the premium calculation module;

[0010] The premium calculation module is configured to, upon receiving an early warning instruction, extract dynamic behavioral features from the insurance business dataset, calculate intermediate indicators such as payment stability and loyalty impact, and obtain the dynamic premium coefficient through multi-feature weighted fusion. Quantify the extent to which the customer's current risk affects the adjustment of the premium;

[0011] The claims assessment module is configured to use machine learning technology to build a claims assessment model, which will include risk coefficients. and dynamic premium coefficient The input model is processed to obtain a dimensionless data set, and then fitted to output a claims assessment index. The assessment results are displayed in visual formats such as heat maps and time-series line graphs to assist in claims decision-making.

[0012] Preferably, the data acquisition module includes a data acquisition unit, a data preprocessing unit, and a feature extraction unit;

[0013] The data acquisition unit is configured to synchronously collect data from an internal database and an external data source: the internal database covers basic policy information, claims history, and basic customer information; the external data source covers economic indicators, weather data, and geographic risk data.

[0014] The data preprocessing unit is configured to execute a three-level data processing flow:

[0015] Data cleaning: Mean filtering is used to process internal policy data, and sliding window filtering is used to process external environmental data;

[0016] Standardization process: through Standardization maps multi-source data to the [0,1] interval, as shown in the formula. ,in This is the original data. These are the minimum and maximum values ​​for this data dimension, respectively, to eliminate the difference in dimensions between different data.

[0017] Anomaly detection: The Isolation Forest algorithm is used to identify anomalies in the preprocessed data, mark the abnormal data, and trigger manual verification.

[0018] The feature extraction unit is configured to extract a multimodal feature set based on the preprocessed data:

[0019] Historical risk characteristics: including average number of claims per year Average annual claim amount This reflects the customer's past risk level;

[0020] Dynamic behavioral characteristics: including customer payment timeliness rate On-time payment count / total payment count, valued between 0 and 1; Customer loyalty. Continuous insurance years / Total insurance years, valued between 0 and 1; Economic volatility index Adjusted based on GDP growth rate, the value ranges from 0.8 to 1.2, with higher growth rates indicating higher values. The closer it is to 1.2, the higher the weather risk index. Based on the annual disaster frequency classification of the insured area, values ​​range from 1 to 5, with 1 representing the lowest risk and 5 representing the highest risk; Geographical risk index. Based on the regional accident rate classification, the value ranges from 1 to 5, with 1 being the lowest risk and 5 being the highest risk.

[0021] Preferably, the risk assessment module includes a historical analysis unit and a risk calculation unit;

[0022] The historical analysis unit is configured to analyze the customer's historical risk status based on historical risk characteristics:

[0023] Calculate the benchmark risk ratio The formula is: ,in The average number of claims per year, This represents the average annual claim amount; the formula reflects the claim frequency corresponding to a unit claim amount. The higher the frequency, the more frequent the customer's risk events, and the higher the baseline risk.

[0024] Calculate customer risk level :

[0025] Step 1.1: Construct a 1-10 point scoring system based on the customer's basic information. Add 2 points for age over / under 60 years old and subtract 1 point for age under 60 years old. Add 3 points for high-risk occupations and subtract 2 points for low-risk occupations. Add 3 points for claims records in the past 3 years and subtract 2 points for no claims records to obtain the risk level score for each customer. ;

[0026] Step 1.2: Calculate the average risk level of the customer group The formula is ,in This refers to the total number of customers who purchased insurance during the same period. This mean is used to eliminate the influence of extreme individual scores and reflects the baseline risk level of the current customer group.

[0027] The risk calculation unit is configured to combine historical risk and group risk to quantify the overall risk of the customer.

[0028] Preferably, the risk calculation unit obtains the risk coefficient through the following steps. :

[0029] Dimensionless processing: for the benchmark risk ratio and customer risk level Z-score standardization is used, and the formula is: ,in This is the average value of the indicator in the historical database. The standard deviation is used to ensure that the two indicators are on the same order of magnitude;

[0030] Weighted fusion computing The formula is ;

[0031] in, This is the standardized baseline risk ratio. The standardized average risk level of customers, The weight values ​​were set using the Analytic Hierarchy Process (AHP) combined with the opinions of five insurance risk control experts. The value ranges from 0.6 to 0.8, highlighting the core impact of historical risk frequency; A value of 0.2-0.4 is used to help correct for differences in group risk. The first correction constant, with a value of 0.01-0.05, is used to correct system biases during the data acquisition process.

[0032] Preferably, the risk calculation unit sets the risk threshold through the following steps. And diagnose high-risk hazards:

[0033] Threshold construction:

[0034] Step 2.1: Collect risk coefficient samples from normal customers over the past 3 years, with a sample size of ≥5000 cases;

[0035] Step 2.2: Use the K-means clustering algorithm to divide the samples into two clusters: low-risk and medium-risk. Take the minimum boundary value of the medium-risk cluster as the initial value. ;

[0036] Step 2.3: Update the sample database quarterly based on newly added normal customer data and recalculate. To ensure the timeliness of the threshold;

[0037] Risk diagnosis:

[0038] like If a customer is identified as having a high risk profile, an early warning instruction is immediately sent to the premium calculation module, and the risk coefficient and abnormal characteristics are recorded simultaneously.

[0039] like If it is determined that there are no high-risk risks for the current customer, the customer information will be classified into the normal risk pool and reassessed every six months.

[0040] Anomaly tracing: If two consecutive assessments It automatically extracts the customer's historical data to generate a risk analysis report, which is then pushed to the risk control specialist.

[0041] Preferably, the premium calculation module includes a behavior analysis unit and a premium adjustment unit;

[0042] The behavior analysis unit is configured to calculate two core intermediate indicators based on dynamic behavior characteristics after receiving an early warning command:

[0043] Payment stability The formula is: ;

[0044] in, Customer payment on-time rate, with a value between 0 and 1; It is a nonlinear adjustment term. The closer it is to 1, the closer this value is to 0.693, indicating a positive impact of enhanced loyalty on payment stability; This is a weather risk index, ranging from 1 to 5. The higher the weather risk, the larger the denominator. The lower the value, the more consistent it is with the actual scenario of payment delays caused by severe weather;

[0045] Value range: The value ranges from 0 to 0.693; a higher value indicates more stable customer payment behavior and lower risk.

[0046] Loyalty Influence The formula is: ;

[0047] in, Customer loyalty is represented by a value between 0 and 1. This is an economic volatility index, ranging from 0.8 to 1.2, indicating a positive economic trend. High, positively moderating the impact of loyalty; This is a geographical risk index, ranging from 1 to 5. The higher the geographical risk, the larger the denominator. The lower the value, the less effective the risk mitigation effect of customer loyalty in high-risk regions becomes;

[0048] Value range: The value ranges from 0 to 0.24, with higher values ​​indicating a stronger mitigating effect of customer loyalty on risk.

[0049] The premium adjustment unit is configured to calculate a dynamic premium coefficient based on intermediate indicators to quantify the premium adjustment range.

[0050] Preferably, the premium adjustment unit obtains the dynamic premium coefficient through the following steps. :

[0051] Indicator standardization: Min-Max normalization is used to map to the [0,1] interval, eliminating magnitude differences;

[0052] Weighted fusion computing The formula is:

[0053] ;

[0054] in, The importance of features was determined through random forest feature importance analysis. With a value of 0.4-0.5, payment stability has the greatest impact on premiums; The value ranges from 0.3 to 0.4, with loyalty having the next highest impact. The value ranges from 0.1 to 0.2, with economic fluctuations as a secondary factor.

[0055] Preferably, the premium adjustment unit diagnoses the premium adjustment range through the following steps:

[0056] Set premium adjustment threshold :

[0057] Step 3.1: Based on the insurance company's profit model, calculate the critical coefficient that can maintain the target profit margin after premium adjustments, as the initial value. ;

[0058] Step 3.2: Based on market competition, assess the initial... Fluctuation range: 10%-15%;

[0059] Step 3.3: Set up customer tiers Individual customers =1.0-1.1, Enterprise Customers =1.1-1.2;

[0060] Adjustment range diagnosis:

[0061] like If the customer's current risk is determined to be low, only a slight adjustment to the premium is needed, or the original premium may be maintained.

[0062] like If the customer is deemed to be at high risk and the premium needs to be significantly increased, a risk warning will be sent to the customer.

[0063] Adjust records: The adjustment range and customer feedback are recorded in the database as the basis for subsequent weight optimization.

[0064] Preferably, the claims assessment module includes a modeling unit and an assessment output unit;

[0065] The modeling unit is configured to construct a claims assessment model through the following steps:

[0066] Data preparation: Using the insurance business dataset of the past 5 years as a sample, the input features are historical risk characteristics and dynamic behavioral characteristics, totaling 7 dimensions: The labels are categorized into claim risk levels from 1 to 5, with 1 representing low risk and 5 representing high risk.

[0067] Model selection and training:

[0068] Select gradient boosting tree As a basic model;

[0069] The system is divided into training, validation, and test sets, with claims accuracy and fraud detection rate as the core indicators. Parameters are optimized using 5-fold cross-validation.

[0070] Model training: Calculate the validation set metrics every 100 iterations. If the metrics do not improve for 20 consecutive iterations, stop training and save the optimal model.

[0071] Model iteration: The model is updated every six months with new claims data and the parameters are retrained to ensure the model's adaptability to new risk patterns;

[0072] The evaluation output unit is configured to output the risk coefficient. Dynamic premium coefficient Input the model, output the claims assessment results.

[0073] Preferably, the assessment output unit calculates the claims assessment index through the following steps. And implement risk response:

[0074] calculate :

[0075] Step 4.1: Calculate the adjustment factor for the dynamic premium coefficient. The formula is: ,in, The base of the natural logarithm; this factor takes values ​​from 0 to 1. The closer to 1, The closer it is to 1, the higher the average annual claim amount. The weights; The closer to 0, The closer the value is to 0.368, the higher the risk factor. The weights;

[0076] Step 4.2: Fitting the output The formula is:

[0077] ;

[0078] in, The average annual claim amount, This is the risk coefficient, ranging from 0 to 1. Multiplying it by 100 is to... Magnitude matching This is the second correction constant, ranging from 0.02 to 0.06, used to correct model prediction errors.

[0079] Value range: The value ranges from 0 to 100, with higher values ​​indicating higher claims risk.

[0080] Set claim thresholds :

[0081] Step 5.1: Collect normal claim cases from the past two years Sample, calculate mean and standard deviation ;

[0082] Step 5.2: For high-risk insurance products, Reduced by 10%;

[0083] Risk Response and Visualization:

[0084] like If a customer's claim risk is determined to be high, an insurance specialist will be immediately notified to conduct a manual review and initiate a fraud investigation procedure.

[0085] like The system determines that the customer's claim risk is in a normal state, automatically processes claim applications, and monitors changes in customer behavior every quarter.

[0086] Visualization: Using heat maps to showcase customers in different regions Distribution, displayed as a line chart for individual customers. The temporal changes can assist management in decision-making.

[0087] This invention provides an insurance business evaluation system based on big data analysis, which has the following beneficial effects:

[0088] 1. The system integrates internal data (policy information, claims history, customer basic information) with external data (economic indicators, weather data, geographical risk data) through the data acquisition module, and extracts a two-dimensional feature set of historical risk characteristics and dynamic behavioral characteristics to form a complete customer risk profile. For example, by combining the customer's average annual claims record (historical characteristics) with the payment timeliness rate and regional disaster risk (dynamic characteristics), the system can more comprehensively reflect the customer's actual risk level, avoid assessment bias caused by single data, and provide a high-quality data foundation for subsequent risk coefficient calculation and premium adjustment.

[0089] 2. The risk assessment module achieves dynamic risk diagnosis by weightedly integrating historical and group risks and combining them with regularly updated risk thresholds. The premium calculation module quantifies the impact of risk on premium adjustments based on dynamic customer behavior characteristics (such as payment stability and loyalty), breaking away from the traditional one-size-fits-all pricing model. For example, for low-risk customers who pay promptly and have long-term insurance policies, premiums can be maintained or slightly adjusted to enhance customer loyalty; for high-risk customers (such as those with frequent claims or payment delays), premiums are promptly increased and risk warnings are pushed to reduce the insurance company's payout risk, achieving a personalized risk control goal of matching risk with premiums.

[0090] 3. The claims assessment module employs the XGBoost machine learning model, combining risk coefficients and dynamic premium coefficients to fit and output a claims assessment index. Through regular model iterations, it adapts to new risk patterns, significantly improving the efficiency and accuracy of claims decisions. Simultaneously, the system displays assessment results through visualizations such as heatmaps and time-series line graphs, assisting management in quickly grasping regional risk distribution and customer risk trends. For high-risk claims (such as fraud), it automatically triggers manual review and investigation procedures, while automating the processing of normal claims (such as rapid disbursement of small claims). This shortens the claims cycle, improves customer experience, effectively identifies fraud risks, and reduces the insurance company's operating costs.

[0091] 4. Each module of the system is designed with a dynamic update mechanism. For example, the risk threshold is optimized every quarter based on new data, and the claims model is iterated every six months in combination with new cases to ensure that the assessment standards are synchronized with changes in market risks. At the same time, the data collection and feature extraction module supports the addition of new data sources (such as vehicle driving data and health monitoring data collected by IoT devices), which can be flexibly extended to multiple insurance scenarios such as auto insurance, health insurance, and property insurance, providing a reusable technical framework for digital risk control in the insurance industry.

[0092] In summary, this invention integrates internal and external multi-source data to construct a complete customer risk profile, enabling dynamic risk assessment and personalized premium adjustments. It leverages machine learning models to improve claims assessment efficiency and fraud detection capabilities. Furthermore, each module has dynamic update and expansion capabilities, effectively solving the problems of low accuracy, poor adaptability, and low efficiency in traditional assessment models. This helps insurance companies achieve precise risk control, optimize customer experience, and reduce operating costs. Attached Figure Description

[0093] Figure 1 This is a schematic diagram of the insurance business evaluation system based on big data analysis as described in this invention.

[0094] Figure 2 This is a schematic diagram of the data acquisition module of the present invention;

[0095] Figure 3 This is a schematic diagram of the risk assessment module of the present invention.

[0096] Figure 4 This is a schematic diagram of the premium calculation module of the present invention.

[0097] Figure 5 This is a schematic diagram of the claims assessment module of the present invention. Detailed Implementation

[0098] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0099] like Figures 1-5As shown, this invention provides a technical solution: an insurance business evaluation system based on big data analysis, including a data acquisition module, a risk assessment module, a premium calculation module, and a claims assessment module. The data acquisition module is configured to collect multi-source insurance-related data (including internal policy data, external environmental data, and customer behavior data), and construct a structured insurance business dataset through data cleaning, standardization, and feature extraction processing to provide a high-quality data foundation for subsequent evaluation. The risk assessment module is configured to analyze customer risk status based on the insurance business dataset: first, calculate a benchmark indicator reflecting the customer's historical risk, and then combine it with the customer group's risk level to obtain a risk coefficient through weighted fusion and dimensionless processing. Preset risk threshold ,Will and If compared, Exceed If a high-risk vulnerability is identified, an early warning instruction is sent to the premium calculation module. Upon receiving the early warning instruction, the premium calculation module is configured to extract dynamic behavioral features (payment timeliness rate, customer loyalty, etc.) from the insurance business dataset, calculate intermediate indicators such as payment stability and loyalty impact, and obtain the dynamic premium coefficient through multi-feature weighted fusion. This quantifies the impact of a customer's current risk on premium adjustments; the claims assessment module is configured to use machine learning technology to build a claims assessment model, incorporating risk coefficients... and dynamic premium coefficient The input model is processed to obtain a dimensionless data set, and then fitted to output a claims assessment index. The assessment results are presented in a visual format, such as heat maps and time-series line graphs, to assist in claims decision-making.

[0100] More specifically, such as Figure 2 As shown, the data acquisition module includes a data acquisition unit, a data preprocessing unit, and a feature extraction unit;

[0101] The data acquisition unit is configured to synchronously collect data from an internal database and external data sources: the internal database covers basic policy information (insured type, sum insured, term), claims history (claim time, amount, reason), and basic customer information (age, occupation, region); the external data sources cover economic indicators (GDP growth rate, inflation rate), weather data (annual disaster frequency and intensity in the insured area), and geographic risk data (regional accident rate, fraud rate).

[0102] The data preprocessing unit is configured to execute a three-level data processing flow:

[0103] Data cleaning: Mean filtering is used to process internal policy data (e.g., removing outliers in claim amounts that deviate from the mean by 3 times the standard deviation, eliminating high-frequency noise), and sliding window filtering is used to process external environmental data (e.g., smoothing the weather risk index with a 7-day window to reduce the impact of short-term fluctuations).

[0104] Standardization process: through Standardization maps multi-source data to the [0,1] interval, as shown in the formula. ,in This is the original data. These are the minimum and maximum values ​​for this data dimension, respectively, to eliminate the difference in dimensions between different data.

[0105] Anomaly detection: The isolated forest algorithm is used to identify anomalies in the preprocessed data (such as a customer's single claim amount far exceeding the average of the same type of insurance), mark the abnormal data, and trigger manual verification.

[0106] The feature extraction unit is configured to extract a multimodal feature set based on the preprocessed data:

[0107] Historical risk characteristics: including average number of claims per year (Total number of claims per year / number of years of insurance coverage), average annual claim amount (Total annual claims / number of years of insurance coverage) reflects the customer's past risk level;

[0108] Dynamic behavioral characteristics: including customer payment timeliness rate On-time payment count / total payment count, valued between 0 and 1; Customer loyalty. Continuous insurance years / Total insurance years, valued between 0 and 1; Economic volatility index Adjusted based on GDP growth rate, the value ranges from 0.8 to 1.2, with higher growth rates indicating higher values. The closer it is to 1.2, the higher the weather risk index. Based on the annual disaster frequency classification of the insured area, values ​​range from 1 to 5, with 1 representing the lowest risk and 5 representing the highest risk; Geographical risk index. Based on the regional accident rate classification, the value ranges from 1 to 5, with 1 being the lowest risk and 5 being the highest risk.

[0109] The data acquisition unit breaks through the limitations of traditional assessments that rely solely on static internal data by simultaneously collecting data from internal databases (basic policy information, claims history, and basic customer information) and external data sources (economic indicators, weather data, and geographic risk data). For example, in auto insurance assessments, combining the annual frequency of heavy rain in the insured area (external weather data) with the customer's occupation (internal customer information) allows for a more accurate capture of the combined impact of regional environmental risks and the customer's own attributes, avoiding misjudgments due to limited data. Internal data covers core insurance business processes, ensuring the authenticity and traceability of the basis for customer risk assessment; external data relates to the macro environment and regional risks, supplementing external factors influencing customer risk. The combination of these two elements constructs a multi-dimensional data pool, providing comprehensive data support for subsequent risk profiling, premium calculation, and claims assessment, solving the "data silo" problem of traditional assessments.

[0110] The data preprocessing unit employs mean filtering for internal policy data to effectively remove outliers in claim amounts that deviate from the mean by three times the standard deviation, eliminating high-frequency noise (such as extreme claim amounts caused by data entry errors). For external environmental data, a 7-day window sliding filter is used to smooth short-term fluctuations in the weather risk index (such as the interference of a single-day sudden disaster on monthly risk assessment), ensuring that the data reflects the true risk trend. The isolated forest algorithm accurately identifies data anomalies after preprocessing (such as a customer's single claim amount far exceeding the average for the same insurance type) and triggers manual verification, preventing abnormal data from entering the subsequent assessment process. For example, it can promptly detect abnormal data related to "false high-amount claims," ​​reducing the interference of fraud risk on the assessment results. Min-Max standardization is used to uniformly map multi-source data to the [0,1] interval, resolving the issue of conflicting dimensions between different data dimensions (such as the difference in magnitude between "claim amount (ten thousand yuan)" and "payment timeliness rate (0-1)"). Standardized data can directly participate in weighted calculations, ensuring the rationality of subsequent calculations of risk coefficients, premium coefficients, and other indicators, avoiding assessment biases caused by dimensional issues.

[0111] The function of the feature extraction unit is as follows: historical risk characteristics (average number of claims per year) Average annual claim amount ): Directly quantify the client's past risk level, for example A higher value indicates a higher frequency of historical risk events for the customer, providing historical reference for risk assessment and addressing the problem of traditional assessments neglecting the inertia of historical customer risk. This also includes the customer's timely payment rate. Loyalty (Reflecting the client's own behavioral risks), and the economic volatility index. Weather risk index Geographical risk index (Reflecting external dynamic risks). For example, (High payment timeliness) combined (High loyalty) reflects low customer performance risk; (High weather risk) combined (High geographical risk) can indicate the cumulative impact of the regional environment on customer risk.

[0112] The multimodal feature set includes both static historical data and dynamically changing behavioral and environmental data, reflecting real-time changes in customer risk. For example, when... When the economic impact factor increases from 0.8 (economic downturn) to 1.2 (economic upturn), the economic impact factor in the calculation of customer premiums can be adjusted in a timely manner, so as to realize the synchronization of risk assessment with changes in the external environment, provide feature support for subsequent personalized premium adjustments and dynamic claims assessment, and solve the limitations of static and fixed traditional assessment.

[0113] More specifically, such as Figure 3 As shown, the risk assessment module includes a historical analysis unit and a risk calculation unit;

[0114] The historical analysis unit is configured to analyze a customer's historical risk profile based on historical risk characteristics.

[0115] Calculate the benchmark risk ratio The formula is: ,in This represents the average number of claims per year. This represents the average annual claim amount (in ten thousand yuan / year). This formula reflects the claim frequency corresponding to a unit claim amount. A higher frequency indicates more frequent customer risk events and a higher baseline risk (e.g., ...). Next / year When the annual income is 10,000 yuan, This means that 0.4 claims correspond to every 10,000 yuan of claim amount.

[0116] Calculate customer risk level :

[0117] Step 1.1: Based on the customer's basic information (age, occupation, past claims history), construct a 1-10 point scoring system. Add 2 points for those over 60 years old and subtract 1 point for those under 60 years old. Add 3 points for high-risk occupations (such as construction workers) and subtract 2 points for low-risk occupations (such as office workers). Add 3 points for claims history in the past 3 years and subtract 2 points for no claims history to obtain the risk level score for each customer. ;

[0118] Step 1.2: Calculate the average risk level of the customer group The formula is ,in The total number of customers who purchased insurance during the same period ( ), This mean is used to eliminate the influence of extreme individual scores and reflects the baseline risk level of the current customer group.

[0119] The risk calculation unit is configured to combine historical risk and group risk to quantify the overall risk of customers.

[0120] The effect of historical analysis units: through formulas The average number of claims per year ( (reflecting the frequency of risk events) and the average annual claim amount ( By combining the severity of risk events with the frequency of claims per unit of claim amount, an indicator called "claim frequency" is constructed to quantify the correlation between the "frequency-amount" relationship of a customer's historical risks. For example, when... =2 times / year When the annual income is 50,000 yuan, =40%, clearly reflecting the risk characteristic of "0.4 claims for every 10,000 yuan of claim amount", avoiding the one-sided risk judgment caused by traditional assessments that only consider the number of claims or the amount.

[0121] For high-frequency, low-amount claims (such as...) =5 times / year =100,000 yuan / year =50%), this indicator can effectively identify the "small amount, high frequency" risk characteristics of such customers; although the amount of a single claim is low, the risk events occur frequently, and the long-term accumulated risk is high. It solves the problem of ignoring high-frequency risks due to "low amount per claim" in traditional assessments, and provides a key basis for subsequent high-risk customer early warning.

[0122] Customer risk level The effects of the calculation: Constructing a standardized individual risk scoring system: Based on basic customer information (age, occupation, past claims history), a 1-10 point scoring system is constructed. Through differentiated scoring rules such as "add 2 points for age 60 and above / subtract 1 point for age below 60," "add 3 points for high-risk occupations / subtract 2 points for low-risk occupations," and "add 3 points for claims history within the past 3 years / subtract 2 points for no history," qualitative customer attributes are transformed into quantitative scores. For example, a 65-year-old construction worker (a high-risk occupation) with a claim record in the past 3 years, single customer rating. =10 points, to achieve standardized comparison of individual risks of different customers, and avoid the subjectivity of "judging customer risk based on experience" in traditional assessment.

[0123] Eliminating the interference of individual extreme values ​​and reflecting the group baseline risk: through the formula Calculating the average risk level of the customer group (M≥1000) can smooth out the impact of extreme individual scores (such as an individual customer's extreme score of 10 or 1 due to special circumstances) on the overall assessment. For example, when a few customers in a group score 10 due to "high-risk occupation + multiple claims," ​​while most customers score 4-6, It can accurately reflect the overall baseline risk level of the group, providing a reference for subsequent "comparison of individual risk and group risk", and solving the assessment bias problem of "individual risk being divorced from the group context" in traditional assessment.

[0124] Effect of the Risk Calculation Unit: The Risk Calculation Unit outputs the "benchmark risk ratio" from the historical analysis unit. (Reflecting the frequency of an individual's historical risks) and "average risk level of a customer group" Using "(reflecting the group's baseline risk)" as input, subsequent dimensionless processing and weighted fusion combine "individual historical risk" with "group reference risk," avoiding the limitations of single-dimensional assessment. For example, a client... Higher (high-frequency individual risk), but The risk level is also relatively high (the overall risk of this group is relatively high). By using weighted fusion, the relative level of the customer's risk in the group can be objectively judged, rather than judging it as high risk based solely on individual data, thus improving the objectivity of risk assessment.

[0125] The risk coefficient is the final output of the risk calculation unit. (Subsequent calculation steps) is the core result of the risk assessment module. This coefficient is derived based on historical and group-based risk quantification and can be directly compared with preset risk thresholds. Comparison to achieve "high-risk hazard identification" (Issuing warnings in a timely manner). For example, when and When the risk level is 0.6, the system can accurately identify the customer as high-risk and promptly send an early warning instruction to the premium calculation module. This provides precise quantitative triggering conditions for subsequent personalized premium adjustments and risk management, solving the problem of "no clear quantitative standard for risk assessment" in traditional assessments.

[0126] More specifically, the risk calculation unit obtains the risk coefficient through the following steps. :

[0127] Dimensionless processing: for the benchmark risk ratio (Percentage form) and customer risk level (1-10 score scale) Z-score standardization is used, and the formula is: ,in This is the average value of the indicator in the historical database. The standard deviation is used to ensure that the two indicators are on the same order of magnitude;

[0128] Weighted fusion computing The formula is ;

[0129] in, This is the standardized baseline risk ratio. The standardized average risk level of customers, The weight values ​​were set using the Analytic Hierarchy Process (AHP) combined with the opinions of five insurance risk control experts. The value ranges from 0.6 to 0.8, highlighting the core impact of historical risk frequency; A value of 0.2-0.4 is used to help correct for differences in group risk. This is the first correction constant, with a value of 0.01-0.05, used to correct system biases during data acquisition.

[0130] Example: If ,but The value ranges from 0 to 1, and the higher the value, the higher the risk.

[0131] More specifically, the risk calculation unit sets the risk threshold through the following steps. And diagnose high-risk hazards:

[0132] Threshold construction:

[0133] Step 2.1: Collect risk coefficient samples from normal customers (no insurance fraud record, average annual claim rate ≤ 5%) over the past 3 years, with a sample size of ≥ 5000 cases;

[0134] Step 2.2: Use the K-means clustering algorithm to divide the samples into two clusters: low-risk and medium-risk. Take the minimum boundary value of the medium-risk cluster as the initial value. ;

[0135] Step 2.3: Update the sample database quarterly based on newly added normal customer data and recalculate. Ensure the timeliness of the threshold (such as the initial threshold). =0.6, adjusted to 0.58 after the quarterly update);

[0136] Risk diagnosis:

[0137] like If a customer is identified as having high-risk potential (such as fraud risk or high-frequency claims risk), an early warning instruction is immediately sent to the premium calculation module, and the risk coefficient and abnormal characteristics (such as...) are recorded simultaneously. (far exceeding the group average)

[0138] like If it is determined that there are no high-risk risks for the current customer, the customer information will be classified into the normal risk pool and reassessed every six months.

[0139] Anomaly tracing: If two consecutive assessments The system automatically extracts the customer's historical data (claims records, payment behavior) to generate a risk analysis report, which is then pushed to the risk control specialist.

[0140] pass Identifying high-risk hazards The binary diagnostic logic, categorized into the normal risk pool, enables a clear classification of customer risks. When Time (e.g.) =0.62, =0.58), the system can quickly identify fraud risks and high-frequency claim risks of customers, immediately send early warning instructions to the premium calculation module, and simultaneously record the risk coefficient and Abnormal characteristics, such as values ​​far exceeding the group average, provide precise trigger signals for subsequent premium increases and risk management; when At the same time, customers are classified into the normal risk pool and reassessed every six months. This reduces excessive intervention in low-risk customers and prevents dynamic changes in risk from being overlooked through regular review, thus balancing risk control accuracy and customer experience.

[0141] "Two consecutive assessments" in the anomaly tracing process The system automatically generates risk analysis reports and assigns them to risk control specialists, creating a closed loop for risk management. Compared to the lack of follow-up after a single high-risk warning in traditional assessments, two consecutive high-risk determinations can eliminate interference from accidental factors (such as misjudgments caused by single data anomalies), ensuring the accuracy of tracing. It automatically extracts customers' historical claim records and payment behavior to generate reports, helping risk control specialists quickly pinpoint the root cause of risks (such as a customer making frequent small claims for three consecutive years, suspected of fraud). This achieves closed-loop management from risk identification to risk tracing to manual intervention, improving the targeting and effectiveness of high-risk customer management and reducing insurance company payout losses.

[0142] More specifically, such as Figure 4 As shown, the premium calculation module includes a behavior analysis unit and a premium adjustment unit;

[0143] The behavior analysis unit is configured to calculate two core intermediate indicators based on dynamic behavior characteristics after receiving an early warning command:

[0144] Payment stability The formula is: ;

[0145] in, The customer payment on-time rate is a value between 0 and 1; (e.g., 9 out of 10 payments by a customer are made on time). ), For nonlinear adjustment term ( The closer the value is to 1, the closer it is to 0.693, indicating a positive impact of increased loyalty on payment stability. This is a weather risk index, ranging from 1 to 5. The higher the weather risk, the larger the denominator. The lower the value, the more consistent it is with the actual scenario of payment delays caused by severe weather;

[0146] Value range: The value ranges from 0 to 0.693; a higher value indicates more stable customer payment behavior and lower risk.

[0147] Loyalty Influence The formula is: ;

[0148] in, Customer loyalty is represented by a value between 0 and 1. This is an economic volatility index, ranging from 0.8 to 1.2, indicating a positive economic trend. High, positively moderating the impact of loyalty; This is a geographical risk index, ranging from 1 to 5. The higher the geographical risk, the larger the denominator. The lower the value, the less effective the risk mitigation effect of customer loyalty in high-risk regions becomes;

[0149] Value range: The value ranges from 0 to 0.24, with higher values ​​indicating a stronger mitigating effect of customer loyalty on risk.

[0150] The premium adjustment unit is configured to calculate the dynamic premium coefficient based on intermediate indicators, thereby quantifying the premium adjustment range.

[0151] formula Customer loyalty ( ) and macroeconomics ( ), regional risks ( This, combined with other factors, allows for a precise measurement of the risk mitigation effect of loyalty. For example, =0.9 (High Loyalty) =1.2 (economic improvement) =1 (extremely low regional risk), calculated as follows This is corrected according to the value range: because ≤1、 ≤1.2、 ≥1, actual ,when When the value is 0.24 (upper limit), it indicates that customer loyalty has the strongest risk mitigation effect under conditions of a favorable economy and low regional risk; when... When the value is 5 (extremely high regional risk), the denominator increases, leading to... The reduction aligns with the reality that loyalty's risk mitigation effect is weakened in high-risk areas, filling the gap in traditional assessments where there are no quantitative standards for the role of loyalty.

[0152] The introduction of the (economic volatility index) allows the risk impact of loyalty to be dynamically adjusted according to the economic environment (such as during economic downturns). =0.8, even high, It will also decrease, reflecting a weakening ability to mitigate the risks of loyalty under economic pressure. The (Geographic Risk Index) reflects the impact of regional differences on loyalty, avoiding the irrationality of a nationally unified loyalty assessment standard and making risk quantification more adaptable to different scenarios.

[0153] The premium adjustment unit serves to provide the core basis for dynamic premium calculation: the output of the behavior analysis unit. (Payment stability) (Loyalty Impact) is the dynamic premium coefficient calculated by the premium adjustment unit. Key intermediate indicators. Through subsequent standardization and weighted fusion, multi-dimensional characteristics of customers, such as payment risk and loyalty risk, can be transformed into a single... It directly quantifies the impact of a customer's current risk on premium adjustments, solving the problem of traditional assessments where premium adjustments rely on experience and lack data support.

[0154] Calculated based on intermediate indicators It can be combined with preset thresholds (such as individual customers). =1.0-1.1) Determine the direction and magnitude of premium adjustments; when When (low risk), the premium is slightly adjusted or remains unchanged; when ≤ At times (high risk), premiums increase significantly. For example, customers... =1.05 ( =1.0), judged as low risk, premium remains unchanged; customer =0.95 ( If the risk level is 1.0, it is considered high risk, and the premium will be increased by 10%-20%. This achieves personalized pricing that matches risk with premium, which not only protects the insurance company's risk control benefits but also avoids excessive payments for low-risk customers, thereby improving customer satisfaction.

[0155] More specifically, the premium adjustment unit obtains the dynamic premium coefficient through the following steps. :

[0156] Indicator standardization: Min-Max normalization is used to map to the [0,1] interval, eliminating magnitude differences;

[0157] Weighted fusion computing The formula is:

[0158] ;

[0159] in, The importance of features was determined through random forest feature importance analysis (trained based on premium adjustment and claims data from the past two years to ensure that the weights reflect the actual risk impact). With a value of 0.4-0.5, payment stability has the greatest impact on premiums; The value ranges from 0.3 to 0.4, with loyalty having the next highest impact. The value ranges from 0.1 to 0.2, with economic fluctuations serving as a secondary factor.

[0160] Example: If ,but The value ranges from 0 to 1. The higher the value, the lower the customer's current risk and the smaller the premium adjustment range.

[0161] Coefficient verification: If the premium adjustment results are compared with those of similar customers in the past, and the error exceeds 5%, the weighting will be re-optimized. .

[0162] More specifically, the premium adjustment unit diagnoses the premium adjustment range through the following steps:

[0163] Set premium adjustment threshold :

[0164] Step 3.1: Based on the insurance company's profit model, calculate the critical coefficient that can maintain the target profit margin after premium adjustments, as the initial value. ;

[0165] Step 3.2: Based on market competition, assess the initial... Fluctuation of 10%-15% (as initially) =1.1, adjusted to 1.05 when competition is intense).

[0166] Step 3.3: Set up customer tiers Individual customers =1.0-1.1, Enterprise Customers =1.1-1.2 (Enterprise customers have more complex risks and higher thresholds).

[0167] Adjustment range diagnosis:

[0168] like If the customer's current risk is determined to be low, the premium only needs to be slightly adjusted (increased by 0%-5%), or the original premium can be maintained.

[0169] like If the customer is deemed to have a high risk profile, the premium needs to be significantly increased (by 10%-20%). At the same time, a risk warning will be sent to the customer (e.g., "Your payment delays have increased, and the premium needs to be increased by 15%").

[0170] Adjust records: The adjustment range and customer feedback are recorded in the database as the basis for subsequent weight optimization.

[0171] pass Determining low risk The logic for determining high risk involves quantifying the magnitude of premium adjustments: low-risk customers receive only slight adjustments (0%-5%) or maintain their original premiums to avoid over-pricing and customer churn; high-risk customers receive substantial increases (10%-20%), using premium premiums to cover potential risk losses. For example, individual customers... The risk level is determined to be low, and the premium remains unchanged; for corporate clients... If a policy is deemed high-risk, the premium will be increased by 15%. This differentiated adjustment addresses the traditional problem of the disconnect between premium risk and price, achieving precise pricing for high-risk, high-premium and low-risk, low-premium policies.

[0172] Send specific risk alerts to high-risk customers (e.g., "Your payment delays have increased, and your premium will need to be increased by 15%)" to make the reasons for premium adjustments transparent. Traditional premium adjustments often raise customer questions due to the lack of clear reasons, while this step explains the basis for the adjustment through specific behavioral characteristics (payment delays), allowing customers to clearly understand the impact of their risky behavior on their premiums and increasing the acceptability of the adjustment. At the same time, it can also guide customers to improve their risky behavior (such as reducing payment delays), forming a virtuous cycle of risk management, behavior improvement, and premium optimization.

[0173] Will The adjustment range and customer feedback records are entered into the database for subsequent weight optimization (such as in the calculation of dynamic premium coefficients). Weights) provide data support. For example, if long-term records show "payment stability ( The impact on claims risk after premium adjustments is greater, but currently... ( The weight is only 0.4, which can be used to adjust the recorded data. The value was increased to 0.45 to improve the model's sensitivity to core risk factors. This closed loop of data recording, weight optimization, and model iteration allows the premium adjustment mechanism to continuously adapt to actual business changes, avoiding assessment biases caused by model rigidity and improving risk control accuracy in the long run.

[0174] More specifically, such as Figure 5 As shown, the claims assessment module includes a modeling unit and an assessment output unit;

[0175] The modeling unit is configured to build a claims assessment model through the following steps:

[0176] Data preparation: Using the insurance business dataset of the past 5 years as a sample, the input features are historical risk characteristics and dynamic behavioral characteristics, totaling 7 dimensions: The labels are categorized into claim risk levels from 1 to 5, with 1 representing low risk and 5 representing high risk.

[0177] Model selection and training:

[0178] Select gradient boosting tree As a basic model (this model has a strong ability to fit nonlinear features and is suitable for insurance risk assessment).

[0179] The dataset is divided into a training set (70%), a validation set (20%), and a test set (10%). The core metrics are claims accuracy and fraud detection rate. 5-fold cross-validation is used to optimize the parameters (learning rate 0.01-0.1, tree depth 3-5, minimum number of samples per leaf node 5).

[0180] Model training: Calculate the validation set metrics every 100 iterations. If the metrics do not improve for 20 consecutive iterations, stop training and save the optimal model.

[0181] Model iteration: The model is updated every six months with new claims data (sample size ≥ 1000 cases), and the parameters are retrained to ensure the model's adaptability to new risk patterns;

[0182] By updating the model and retraining its parameters every six months with newly added claims data (sample size ≥ 1000 cases), new risk patterns in the insurance industry can be captured in a timely manner (such as new fraud methods and changes in claims risks caused by sudden public events). For example, when a new fraud pattern emerges in a certain type of insurance, using false weather disaster claims as evidence, the new data will contain the characteristics of such cases. Through model iteration, the model can learn and identify this type of risk, avoiding the omission of new risks due to model rigidity, ensuring that the model remains adaptable to current risks in the long term, and overcoming the limitation of traditional models that are trained once and remain unchanged for a long time and cannot cope with new risks.

[0183] The evaluation output unit is configured to output the risk coefficient. Dynamic premium coefficient Input the model, output the claims assessment results.

[0184] The core result of the risk assessment module (risk coefficient) The core result of the premium calculation module (dynamic premium coefficient) Input the model to achieve a comprehensive assessment of historical risk and current behavioral risk. Reflects the client's long-term risk level. This reflects the customer's current dynamic risk level. Combining these two indicators provides a comprehensive assessment of the customer's claims risk, avoiding the one-sidedness of assessments based on a single indicator (such as looking only at...). It's possible to overlook recent improvements in customer behavior and only look at... (This may overlook the customer's historical high-risk inertia), providing accurate quantitative basis for claims decisions.

[0185] Supporting differentiated claims decisions: Output claims assessment results (such as claims assessment indices in subsequent steps) This can be directly used to differentiate claim risk levels, thereby guiding differentiated decision-making—high-risk cases trigger manual review and fraud investigation, while low-risk cases are processed automatically. For example, if a customer's claim risk level is determined to be level 5 (high risk) based on the assessment results, manual review is triggered; if it is determined to be level 2 (low risk), automatic and rapid payment is made. This improves the accuracy of risk control for high-risk cases and the efficiency of claims processing for low-risk cases, balancing the insurance company's risk management and customer experience.

[0186] More specifically, the assessment output unit calculates the claims assessment index through the following steps. And implement risk response:

[0187] calculate :

[0188] Step 4.1: Dynamically adjust the weights of risk indicators to suit the client's current risk status:

[0189] Through formula Calculate the adjustment factor to achieve the risk coefficient ( ) and average annual claim amount ( Dynamic weight adaptation. When When the value approaches 1 (the customer's current dynamic risk is low, such as stable payments and high loyalty), Approaching 1, significantly enhanced The weighting; at this time, the customer's current behavior risk is low, and the historical claim amount ( It better reflects the true risk of claims; when When the value is close to 0 (the customer's current dynamic risk is high, such as payment delays or decreased loyalty), Approaching 0.368, a significant enhancement. The weighting; at this point, the customer's current behavioral risk has become apparent, and their long-term risk level should be the primary consideration. This dynamic adjustment mechanism avoids assessment bias caused by fixed weights, making risk quantification more closely match the client's real-time risk status and solving the problem of traditional assessment weights being rigid and unable to flexibly adapt to risk changes.

[0190] Step 4.2: Accurately calculate and manage claims risks to achieve risk quantification and standardization:

[0191] Matching magnitudes ensures the rationality of calculations: (Value 0-1) multiplied by 100, can be compared with (At the ten thousand yuan level) Achieve scale matching (e.g.) When the value is 0.5, multiplying by 100 results in 50, which is the same as... =10, 20, etc. are in the same order of magnitude, so as to avoid the weakening of a certain indicator in the calculation due to the difference in magnitude (e.g., =0.5 directly with =10 added together The impact is almost negligible), ensuring and All can participate effectively Fitting.

[0192] Improving prediction accuracy with a correction constant: Introducing a second correction constant (0.02-0.06), can correct model prediction errors (such as fitting bias caused by data noise or missing features), allowing It more closely reflects the actual claims risks faced by customers. For example, the actual claims risks faced by a customer correspond to... The value should be 55; the model fit value is 54.95. =0.05 can be corrected to 55 to improve the accuracy of the evaluation.

[0193] Standardized values ​​facilitate risk classification: The value ranges from 0 to 100, with higher values ​​indicating higher risk. This standardized range facilitates the intuitive classification of risk levels (e.g., 0-30 for low risk, 31-70 for medium risk, and 71-100 for high risk), providing a clear quantitative basis for subsequent risk response and avoiding the problems of vague risk levels and lack of clear numerical standards in traditional risk assessments.

[0194] Set claim thresholds Steps:

[0195] Step 5.1: Establish benchmarks based on normal cases to ensure the objectivity of the thresholds:

[0196] Collect normal claim cases (no fraud, in compliance with insurance terms) from the past two years. Sample, calculate mean and standard deviation This approach determines the distribution range of normal risks from a data perspective, avoiding reliance on subjective experience in threshold setting (such as setting high-risk thresholds based on intuition in traditional assessments). A sample period of nearly two years covers recent business risk characteristics, ensuring that benchmark data matches current business operations and providing an objective and reliable basis for threshold calculation.

[0197] Step 5.2: Adapting tiered thresholds to different insurance types to enhance risk sensitivity:

[0198] A universal threshold covers most scenarios: The settings can cover 95% of normal cases, meaning that only 5% of normal cases may be misjudged as high-risk, minimizing the possibility of normal cases triggering manual review, and balancing risk control accuracy and claims efficiency.

[0199] Thresholds for high-risk insurance products will be lowered to strengthen risk management: For high-risk insurance products such as property insurance and engineering insurance, the threshold will be lowered by 10% (as originally...). =60, lowered to 54), which can reduce the threshold for identifying high-risk cases; even Even if the risk level is slightly below the general threshold, it can still be judged as high risk and a strict review will be initiated. This meets the business needs of high-risk insurance products, which require more sensitive risk identification. It avoids missing high-risk cases due to excessively high thresholds and reduces the insurance company's payout losses.

[0200] Steps for risk response and visualization:

[0201] Precise management and control of high-risk cases: In such cases, a specialist should be immediately notified to conduct a manual review and initiate a fraud investigation (such as verifying the authenticity of claim materials and obtaining accident evidence). This can accurately identify fraud risks (such as false claims and duplicate claims), prevent insurance companies from incurring unnecessary payouts due to fraud cases, and improve the effectiveness of risk control.

[0202] Efficient handling of normal cases: In the event of a dispute, claims are processed automatically (e.g., small claims are credited within 24 hours), significantly shortening the claims processing cycle for normal cases and preventing customer dissatisfaction due to long waiting times for manual review, thus improving customer experience. Customer behavior changes are monitored quarterly, allowing for timely capture of dynamic adjustments in customer risk (e.g., payment delays for previously low-risk customers), enabling continuous risk management.

[0203] Visualization: Using heat maps to showcase customers in different regions Distribution, displayed as a line chart for individual customers. The temporal changes (over the past 12 months) help management make decisions.

[0204] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An insurance business evaluation system based on big data analysis, characterized in that, It includes a data acquisition module, a risk assessment module, a premium calculation module, and a claims assessment module; The data acquisition module is configured to collect insurance-related multi-source data, and construct a structured insurance business dataset through data cleaning, standardization and feature extraction processing to provide a high-quality data foundation for subsequent evaluation; The risk assessment module is configured to analyze customer risk profiles based on insurance business datasets: first, it calculates benchmark indicators reflecting a customer's historical risk; then, it combines these with the risk levels of the customer group, and obtains risk coefficients through weighted fusion and dimensionless processing. Preset risk threshold ,Will and If compared, Exceed If a high-risk hazard is identified, an early warning instruction is sent to the premium calculation module. The premium calculation module is configured to, upon receiving an early warning instruction, extract dynamic behavioral features from the insurance business dataset, calculate intermediate indicators such as payment stability and loyalty impact, and obtain the dynamic premium coefficient through multi-feature weighted fusion. Quantify the extent to which the customer's current risk affects the adjustment of the premium; The claims assessment module is configured to use machine learning technology to build a claims assessment model, which will include risk coefficients. and dynamic premium coefficient The input model is processed to obtain a dimensionless data set, and then fitted to output a claims assessment index. The assessment results are presented in a visual format, such as heat maps and time-series line graphs, to assist in claims decision-making.

2. The insurance business evaluation system based on big data analysis according to claim 1, characterized in that, The data acquisition module includes a data acquisition unit, a data preprocessing unit, and a feature extraction unit; The data acquisition unit is configured to synchronously collect data from an internal database and an external data source: the internal database covers basic policy information, claim history, and basic customer information; External data sources include economic indicators, weather data, and geographic risk data. The data preprocessing unit is configured to execute a three-level data processing flow: Data cleaning: Mean filtering is used to process internal policy data, and sliding window filtering is used to process external environmental data; Standardization process: through Standardization maps multi-source data to the [0,1] interval, as shown in the formula. ,in The original data, These are the minimum and maximum values ​​for this data dimension, respectively, to eliminate the difference in dimensions between different data. Anomaly detection: The Isolation Forest algorithm is used to identify anomalies in the preprocessed data, mark the abnormal data, and trigger manual verification. The feature extraction unit is configured to extract a multimodal feature set based on the preprocessed data: Historical risk characteristics: including average number of claims per year Average annual claim amount This reflects the customer's past risk level; Dynamic behavioral characteristics: including customer payment timeliness rate On-time payment count / total payment count, valued between 0 and 1; Customer loyalty. Continuous insurance years / Total insurance years, valued between 0 and 1; Economic volatility index Adjusted based on GDP growth rate, the value ranges from 0.8 to 1.2, with higher growth rates indicating higher values. The closer it is to 1.2, the higher the weather risk index. Based on the annual disaster frequency classification of the insured area, values ​​range from 1 to 5, with 1 representing the lowest risk and 5 representing the highest risk; Geographical risk index. Based on the regional accident rate classification, the value ranges from 1 to 5, with 1 being the lowest risk and 5 being the highest risk.

3. The insurance business evaluation system based on big data analysis according to claim 2, characterized in that, The risk assessment module includes a historical analysis unit and a risk calculation unit; The historical analysis unit is configured to analyze the customer's historical risk status based on historical risk characteristics: Calculate the benchmark risk ratio The formula is: ,in The average number of claims per year, This represents the average annual claim amount. This formula reflects the frequency of claims corresponding to a unit claim amount. The higher the frequency, the more frequent the customer's risk events are, and the higher the baseline risk. Calculate customer risk level : Step 1.1: Construct a 1-10 point scoring system based on the customer's basic information. Add 2 points for age over / under 60 years old and subtract 1 point for age under 60 years old. Add 3 points for high-risk occupations and subtract 2 points for low-risk occupations. Add 3 points for claims records in the past 3 years and subtract 2 points for no claims records to obtain the risk level score for each customer. ; Step 1.2: Calculate the average risk level of the customer group The formula is ,in This refers to the total number of customers who purchased insurance during the same period. This mean is used to eliminate the influence of extreme individual scores and reflects the baseline risk level of the current customer group. The risk calculation unit is configured to combine historical risk and group risk to quantify the overall risk of the customer.

4. The insurance business evaluation system based on big data analysis according to claim 3, characterized in that, The risk calculation unit obtains the risk coefficient through the following steps. : Dimensionless processing: for the benchmark risk ratio and customer risk level Z-score standardization is used, and the formula is: ,in This is the average value of the indicator in the historical database. The standard deviation is used to ensure that the two indicators are on the same order of magnitude; Weighted fusion computing The formula is ; in, This is the standardized baseline risk ratio. The standardized average risk level of customers, The weight values ​​were set using the Analytic Hierarchy Process (AHP) combined with the opinions of five insurance risk control experts. The value ranges from 0.6 to 0.8, highlighting the core impact of historical risk frequency; A value of 0.2-0.4 is used to help correct for differences in group risk. The first correction constant, with a value of 0.01-0.05, is used to correct system biases during the data acquisition process.

5. The insurance business evaluation system based on big data analysis according to claim 4, characterized in that, The risk calculation unit sets the risk threshold through the following steps. And diagnose high-risk hazards: Threshold construction: Step 2.1: Collect risk coefficient samples from normal customers over the past 3 years, with a sample size of ≥5000 cases; Step 2.2: Use the K-means clustering algorithm to divide the samples into two clusters: low-risk and medium-risk. Take the minimum boundary value of the medium-risk cluster as the initial value. ; Step 2.3: Update the sample database quarterly based on newly added normal customer data and recalculate. To ensure the timeliness of the threshold; Risk diagnosis: like If a customer is identified as having a high risk profile, an early warning instruction is immediately sent to the premium calculation module, and the risk coefficient and abnormal characteristics are recorded simultaneously. like If it is determined that there are no high-risk risks for the current customer, the customer information will be classified into the normal risk pool and reassessed every six months. Anomaly tracing: If two consecutive assessments It automatically extracts the customer's historical data to generate a risk analysis report, which is then pushed to the risk control specialist.

6. The insurance business evaluation system based on big data analysis according to claim 5, characterized in that, The premium calculation module includes a behavior analysis unit and a premium adjustment unit; The behavior analysis unit is configured to calculate two core intermediate indicators based on dynamic behavior characteristics after receiving an early warning command: Payment stability The formula is: ; in, The customer payment on-time rate, with a value between 0 and 1; It is a nonlinear adjustment term. The closer it is to 1, the closer this value is to 0.693, indicating a positive impact of enhanced loyalty on payment stability; This is a weather risk index, ranging from 1 to 5. The higher the weather risk, the larger the denominator. The lower the value, the more consistent it is with the actual scenario of payment delays caused by severe weather; Value range: The value ranges from 0 to 0.693; a higher value indicates more stable customer payment behavior and lower risk. Loyalty Influence The formula is: ; in, Customer loyalty is represented by a value between 0 and 1. This is an economic volatility index, ranging from 0.8 to 1.2, indicating a positive economic trend. High, positively moderating the impact of loyalty; This is a geographical risk index, ranging from 1 to 5. The higher the geographical risk, the larger the denominator. The lower the value, the less effective the risk mitigation effect of customer loyalty in high-risk regions becomes; Value range: The value ranges from 0 to 0.24, with higher values ​​indicating a stronger mitigating effect of customer loyalty on risk. The premium adjustment unit is configured to calculate a dynamic premium coefficient based on intermediate indicators to quantify the premium adjustment range.

7. The insurance business evaluation system based on big data analysis according to claim 6, characterized in that, The premium adjustment unit obtains the dynamic premium coefficient through the following steps. : Indicator standardization: Min-Max normalization is used to map the data to the [0,1] interval, thus eliminating the difference in magnitude. Weighted fusion computing The formula is: ； in, The importance of features was determined through random forest feature importance analysis. With a value of 0.4-0.5, payment stability has the greatest impact on premiums; The value ranges from 0.3 to 0.4, with loyalty having the next highest impact. The value ranges from 0.1 to 0.2, with economic fluctuations as a secondary factor.

8. The insurance business evaluation system based on big data analysis according to claim 7, characterized in that, The premium adjustment unit diagnoses the premium adjustment range through the following steps: Set premium adjustment threshold : Step 3.1: Based on the insurance company's profit model, calculate the critical coefficient that can maintain the target profit margin after premium adjustments, as the initial value. ; Step 3.2: Based on market competition, assess the initial... Fluctuation range: 10%-15%; Step 3.3: Set up customer tiers Individual customers =1.0-1.1, Enterprise Customers =1.1-1.2; Adjustment range diagnosis: like If the customer's current risk is determined to be low, only a slight adjustment to the premium is needed, or the original premium may be maintained. like If the customer is deemed to be at high risk and the premium needs to be significantly increased, a risk warning will be sent to the customer. Adjust records: The adjustment range and customer feedback are recorded in the database as the basis for subsequent weight optimization.

9. The insurance business evaluation system based on big data analysis according to claim 8, characterized in that, The claims assessment module includes a modeling unit and an assessment output unit; The modeling unit is configured to construct a claims assessment model through the following steps: Data preparation: Using the insurance business dataset of the past 5 years as a sample, the input features are historical risk characteristics and dynamic behavioral characteristics, totaling 7 dimensions: The labels are categorized into claim risk levels from 1 to 5, with 1 representing low risk and 5 representing high risk. Model selection and training: Select gradient boosting tree As a basic model; The system is divided into training, validation, and test sets, with claims accuracy and fraud detection rate as the core indicators. Parameters are optimized using 5-fold cross-validation. Model training: Calculate the validation set metrics every 100 iterations. If the metrics do not improve for 20 consecutive iterations, stop training and save the optimal model. Model iteration: The model is updated every six months with new claims data and the parameters are retrained to ensure the model's adaptability to new risk patterns; The evaluation output unit is configured to output the risk coefficient. Dynamic premium coefficient Input the model, output the claims assessment results.

10. The insurance business evaluation system based on big data analysis according to claim 9, characterized in that, The assessment output unit calculates the claims assessment index through the following steps. And implement risk response: calculate : Step 4.1: Calculate the adjustment factor for the dynamic premium coefficient. The formula is: ,in, The base of the natural logarithm; this factor takes values ​​from 0 to 1. The closer to 1, The closer it is to 1, the higher the average annual claim amount. The weights; The closer to 0, The closer the value is to 0.368, the higher the risk factor. The weights; Step 4.2: Fitting the output The formula is: ； in, The average annual claim amount, This is the risk coefficient, ranging from 0 to 1. Multiplying it by 100 is to... Scale matching, This is the second correction constant, ranging from 0.02 to 0.06, used to correct model prediction errors. Value range: The value ranges from 0 to 100, with higher values ​​indicating higher claims risk. Set claim thresholds : Step 5.1: Collect normal claim cases from the past two years Sample, calculate mean and standard deviation ; Step 5.2: For high-risk insurance products, Reduced by 10%; Risk Response and Visualization: like If a customer's claim risk is determined to be high, an insurance specialist will be immediately notified to conduct a manual review and initiate a fraud investigation procedure. like The system determines that the customer's claim risk is in a normal state, automatically processes claim applications, and monitors changes in customer behavior every quarter. Visualization: Using heat maps to showcase customers in different regions Distribution, displayed as a line chart for individual customers. The temporal changes can assist management in decision-making.

Images