Machine learning-based immigrant management method, system, equipment and medium

Abstract

The invention discloses an immigrant management method, system and device based on machine learning and a medium, and relates to the technical field of immigrant management, and the method comprises the specific steps: building a multi-source data collection system to collect immigrant information data, constructing an immigrant information data set, employing a data flow automation platform to carry out the preprocessing of the immigrant information data set, and carrying out the processing of the immigrant information data set. Constructing a feature importance calculation formula, screening out high-value features, combining a machine learning algorithm, establishing a default risk assessment model and training, outputting an immigrant credit scoring result, and after the default risk assessment model is online, assessing the performance of the default risk assessment model and establishing a feedback mechanism to update the default risk assessment model. The method can quantify the immigrant credit risk and output the immigrant comprehensive credit score, provides reference for immigrant management decision, and can be widely applied to the fields of immigrant credit approval, immigrant settlement management and the like.

Description

Technical Field

[0001] This invention relates to the field of immigration management technology, and in particular to an immigration management method, system, device and medium based on machine learning. Background Technology

[0002] With the continuous development of hydropower projects, a large number of reservoir migrants have been formed. The resettlement process has generated a wealth of data, including basic personal information, family structure, social relationships, and economic activities. Effective social management and services for these specific groups require the integration, analysis, and application of this large amount of dispersed information.

[0003] Current technologies primarily rely on bank credit data and limited demographic information. This data integration is limited in scope, neglecting crucial dimensions specific to immigrant groups, such as relocation and resettlement data, compensation amounts, and resettlement methods. This results in a feature omission rate exceeding 10%, failing to comprehensively reflect the immigrants' true creditworthiness and repayment ability. Existing immigration management scoring models lack dynamic tracking mechanisms, failing to capture dynamic characteristics such as changes in economic conditions, employment transitions, and income fluctuations during the resettlement process. This leads to a risk assessment bias rate as high as 15%–20% within the resettlement period, impacting the accuracy of financial institutions' credit decisions. Furthermore, existing immigration management scoring models employ black-box algorithms or complex nonlinear models, making it difficult for financial institutions and immigrants to understand the scoring criteria, hindering transparency in the credit approval process and the protection of immigrants' rights. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a machine learning-based immigration management method includes: Establish a multi-source data collection system to collect immigration information data and construct an immigration information dataset; Based on the immigration information dataset, an automated data flow platform was used to preprocess the immigration information dataset to obtain a set of key features of immigration information. Based on the key feature set of immigration information, a feature importance calculation formula is constructed, high-value features are screened out, and a feature importance ranking table is output. Based on the feature importance ranking table and machine learning algorithm, a default risk assessment model is established and trained to output immigrant credit score results. After the default risk assessment model is launched, a group stability calculation formula is established to calculate the performance of the default risk assessment model, and a feedback mechanism is established to update the default risk assessment model.

[0006] As a preferred embodiment of the machine learning-based immigration management method described in this invention, the method includes: establishing a multi-source data collection system to collect immigration information data and constructing an immigration information dataset, comprising: Database connection pooling technology is used to collect basic immigration information; We use API interfaces to connect with bank credit reporting centers and third-party credit reporting platforms to collect immigration credit data. Web crawling technology was used to collect data on the social behavior characteristics of immigrants; Collect immigrant socioeconomic data using a government data exchange platform; Establish a data integrity monitoring and feedback mechanism to perform integrity checks on the collected data, and trigger the feedback mechanism when a data integrity problem is detected; An immigrant information dataset is constructed based on the immigrant basic information, immigrant credit data, immigrant social behavior characteristics data, and immigrant socioeconomic data that have passed integrity testing.

[0007] The beneficial effects of this preferred technical solution are as follows: This invention integrates immigrant basic information, credit data, social behavior data, and socioeconomic data through a multi-source data collection system to construct an immigrant information dataset. This expands the data dimensions from a limited number in traditional models to multiple dimensions, reducing the feature missing rate and thus comprehensively reflecting the immigrants' true credit status and repayment ability.

[0008] As a preferred embodiment of the machine learning-based immigration management method described in this invention, the method includes: constructing a feature importance calculation formula based on a key feature set of immigration information, selecting high-value features, and outputting a feature importance ranking table, including: Based on the key feature set of immigration information, the chi-square binning algorithm is used to calculate the chi-square value of adjacent bins, and output a binning scheme corresponding to each feature. Based on the binning characteristics, construct the evidence weight calculation formula and output the evidence weight encoded dataset; Based on the evidence weight encoding dataset, a formula for calculating feature importance is constructed, high-value features are selected, and a feature importance ranking table is output.

[0009] The beneficial effects of this preferred technical solution are as follows: This invention discretizes continuous features into business-meaning intervals by performing chi-square binning, avoiding the oversensitivity of default risk assessment models to extreme values ​​in continuous features. At the same time, it can capture the nonlinear relationship between features and target variables. The feature importance calculation formula provides a specific numerical value for the predictive ability of each feature. Based on the feature importance standard, high-value features can be automatically screened.

[0010] As a preferred embodiment of the machine learning-based immigration management method described in this invention, the method involves: establishing and training a default risk assessment model based on a feature importance ranking table combined with a machine learning algorithm, and outputting an immigration credit score, including: Based on a feature importance ranking table and machine learning algorithms, a default risk assessment model is established, with the following expression: , in, For the probability of default, For the intercept term, is the total number of features, and 'a' is the current feature index. Let be the WOE value of the a-th feature; The default risk assessment model is trained based on the key feature set of immigration information. Based on the training process of the default risk assessment model, a loss function is established to optimize the default risk assessment model, and the expression is: , in, Represents the loss function. For the sample size, For real labels, To predict probabilities, The regularization coefficient is . represents the squared value of the weight of the a-th feature, h represents the total number of features, e represents the index of the current sample, and a represents the index of the current feature; The default risk assessment model parameters are iteratively updated using the gradient descent method until the loss function reaches the maximum number of iterations, and the trained default risk assessment model is then output. Based on the trained default risk assessment model, a scoring conversion formula is established to convert the default probability output by the default risk assessment model into a credit score. The expression is as follows: , , , , in, For the final credit score, Using 600 as the baseline score, derived through data experiments, the value was determined. Let the weight of the a-th feature be , Let be the WOE value of the 'a'-th feature, and 'odds' represent the proportion of good and bad customers in the current sample. This represents the baseline value for the ratio of superior to inferior customers. A represents the offset coefficient in the rating conversion formula, and B represents the scaling coefficient in the rating conversion formula. Both coefficients A and B are calculated. This represents the doubling score. Based on data experiments, a value of 20 indicates that when odds double, the credit score increases by PDO points. is the evidence weight for the baseline bin, and 'a' is the current feature index.

[0011] The beneficial effects of this preferred technical solution are as follows: This invention establishes a default risk assessment model through machine learning algorithms. The magnitude and sign of the weight coefficient of each feature directly reflect the direction and degree of contribution of that feature to the final default probability. The default probability output by the default risk assessment model is mapped to a fixed score range through a scoring conversion formula. The generated score is in a language that financial institutions generally understand and use, and can be directly integrated into the existing approval process.

[0012] Secondly, the present invention provides a machine learning-based immigration management system, including a data acquisition module, a preprocessing module, a feature calculation module, an output module, and a feedback module. The acquisition module is used to establish a multi-source data acquisition system to collect immigration information data and construct an immigration information dataset. The preprocessing module is used to preprocess the immigration information dataset using a data flow automation platform to obtain a set of key features of immigration information. The feature calculation module is used to construct a feature importance calculation formula based on the key feature set of immigration information, screen out high-value features, and output a feature importance ranking table. The output module is used to establish and train a default risk assessment model based on a feature importance ranking table and a machine learning algorithm, and output the immigrant credit score result. The feedback module is used to calculate the performance of the default risk assessment model by establishing a group stability calculation formula after the default risk assessment model is launched, and to establish a feedback mechanism to update the default risk assessment model.

[0013] Thirdly, a computer device includes a memory and a processor, the memory storing a computer program, characterized in that the processor executes the computer program to implement steps of a machine learning-based immigration management method.

[0014] Fourthly, a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of a machine learning-based immigration management method.

[0015] Compared with existing technologies, the advantages of this invention are as follows: In immigrant credit risk assessment methods, traditional methods mainly rely on static bank credit data and limited demographic information, resulting in a single assessment dimension, poor model interpretability, and an inability to capture dynamic changes during the immigrant resettlement process. This invention constructs a dynamic scoring system based on multi-source data fusion and an interpretable machine learning model. It correlates multi-dimensional dynamic characteristics of immigrants, such as resettlement information, compensation data, social behavior, and economic status, with their credit risk to output quantitative assessment results, avoiding the drawbacks of traditional solutions that rely on static snapshot data and repeated manual verification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the overall process of an immigration management method based on machine learning according to an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0019] Example 1, referring to Figure 1 As one embodiment of the present invention, a machine learning-based immigration management method is provided, comprising: It should be noted that as the resettlement of hydropower migrants progresses, the dimensions of migrant credit risk assessment are becoming increasingly complex. Migrant credit assessment requires the integration of multi-source heterogeneous data to construct an assessment model that can reflect the true creditworthiness of migrants. However, this multi-source fusion assessment method involves complex feature engineering and modeling techniques. Given the lack of unified quantitative standards and dynamic update mechanisms in existing technologies, migrant credit assessments often rely on traditional logistic regression models, human experience-based judgments, or complex black-box machine learning models. This necessitates repeated data collection and consumes significant resources for model maintenance and validation, leading to delayed credit decisions, inefficient migrant management, and severe misallocation of financial resources.

[0020] Therefore, addressing the lack of unified multi-source data quantification standards, poor interpretability, and insufficient dynamism in existing technologies, the following steps (S1-S5) are implemented: establishing a multi-source data collection system to collect immigration information data; constructing an immigration information dataset and preprocessing it to obtain a key feature set of immigration information, thus solving the problem of the lack of unified multi-source data quantification standards in existing technologies; constructing an immigration profile vector; constructing a feature importance calculation formula; outputting a feature importance ranking table; combining machine learning algorithms to establish and train a default risk assessment model; outputting immigration credit scores; after the default risk assessment model is launched, establishing a group stability calculation formula to calculate the performance of the default risk assessment model and establishing a feedback mechanism to update the default risk assessment model, thus solving the problems of poor interpretability and insufficient dynamism in traditional assessment models.

[0021] S1: Establish a multi-source data collection system to collect immigration information data and construct an immigration information dataset; S2: Based on the immigration information dataset, the immigration information dataset is preprocessed using a data flow automation platform to obtain a set of key features of immigration information. S3: Based on the key feature set of immigration information, construct a feature importance calculation formula, screen out high-value features, and output a feature importance ranking table; S4: Based on the feature importance ranking table and machine learning algorithm, establish and train a default risk assessment model, and output the immigration credit score result. S5: After the default risk assessment model is launched, establish a group stability calculation formula to calculate the performance of the default risk assessment model and establish a feedback mechanism to update the default risk assessment model.

[0022] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, a machine learning-based immigration management method is provided.

[0023] In step S1, a multi-source data collection system is established to collect immigration information data and construct an immigration information dataset, including the following steps A1-A6: A1: Use database connection pooling technology to collect basic immigration information.

[0024] By using database connection pool technology and the HikariCP connection pool management method, a secure connection channel is established with the hydropower resettlement management system. SQL query statements are used to extract basic resettlement information, including resettlement number, name, ID number, home address, relocation time, resettlement method, compensation amount, amount already paid, and amount to be paid, from specified fields.

[0025] The retrieved immigration basic information is stored in a temporary buffer in JSON format.

[0026] A2: Use API interfaces to connect with bank credit reporting centers and third-party credit reporting platforms to collect immigration credit data.

[0027] Connect to third-party credit reporting platforms such as bank credit reporting centers and Sesame Credit through RESTful API interfaces.

[0028] The immigrant authorizes a credit inquiry through the immigration service platform. After obtaining the authorization token, the system sends an inquiry request to the credit API, transmitting the ID number and the inquiry time range, and receives credit report data, including the number of credit cards, loan records, number of overdue payments, overdue amount, longest overdue days, number of inquiries, etc.

[0029] A3: Use web crawling technology to collect data on immigrant social behavior characteristics.

[0030] Immigrants bind their social media accounts to the immigration service platform and authorize data collection. The web crawler system, based on the authorized scope, crawls user posting content, comments, likes, reposts, and other behavioral data from the past 6 months. Natural language processing technology is used to perform sentiment analysis on the text content to identify implicit information such as employment status, income changes, and consumption tendencies, thereby obtaining immigrant social behavior characteristic data including posting frequency, activity level, sentiment tendencies, and employment keywords.

[0031] A4: Collect immigrant socio-economic data using a government data exchange platform.

[0032] Sign data sharing agreements with local civil affairs departments, human resources and social security departments, and statistics departments to clarify the data scope, update frequency, and security responsibilities. Submit data request lists through the API interface of the government data exchange platform to receive immigrant socio-economic data, including information on minimum living allowance, employment registration, social security payment records, and housing provident fund payment records.

[0033] A5: Establish a data integrity monitoring and feedback mechanism to perform integrity checks on the collected data, and trigger the feedback mechanism when a data integrity problem is detected.

[0034] A6: Construct an immigrant information dataset based on the immigrant basic information, immigrant credit data, immigrant social behavior characteristics data, and immigrant socioeconomic data that have passed the integrity check.

[0035] In this embodiment of the application, the specific steps for establishing the data integrity monitoring and feedback mechanism in step A5 are as follows: Based on the collected basic immigrant information, immigrant credit data, immigrant social behavior characteristics data, and immigrant socioeconomic data, a data integrity monitoring and feedback mechanism will be established.

[0036] The data integrity monitoring includes calculating the field missing rate of each data source, expressed as: Missing rate = (Number of missing fields / Total number of fields) × 100%, Set a missing rate threshold. When the missing rate of a single data source is greater than 5%, an alert is triggered. When critical fields, such as ID card number, immigration number, or compensation amount, are missing, the system immediately marks them as high-priority anomalies.

[0037] The feedback mechanism includes generating an abnormal data report when a data integrity issue is detected, which includes the missing field names, the number of missing fields, and a list of affected immigration numbers. A notification is sent to the data source administrator, requesting that the data be supplemented within 48 hours. For data that fails to be collected by a third-party API, a retry mechanism is initiated, and the abnormal record is stored in the data quality log table for subsequent analysis and optimization.

[0038] In an optional implementation, the data integrity monitoring and feedback mechanism established in step A5 can also employ a real-time detection and alarm method based on a rule engine. In the rule engine, a set of integrity rules is defined. After data collection, it does not first fall into a buffer but directly enters the stream processing pipeline. The system performs rule matching on each record or each micro-batch of data in real time. Once a rule is triggered, the system immediately executes a preset action.

[0039] In another alternative implementation, the data integrity monitoring and feedback mechanism established in step A5 can also adopt a batch processing monitoring scheme based on data quality tools. Data integrity rules are pre-configured in the data quality tools or database. Through scheduling tools or database jobs, after the daily data processing tasks are completed, the data integrity rule check SQL script is automatically executed and a quality report is generated.

[0040] It should be noted that, compared with the prior art, the present invention ensures the integrity of data during the data acquisition stage by establishing a data integrity monitoring and automated feedback mechanism, automates the data governance process through automated monitoring and real-time feedback, and enables the system to have the ability to self-repair and continuously optimize through triggering retry and feedback mechanisms.

[0041] In step S2, based on the immigration information dataset, the immigration information dataset is preprocessed using a data flow automation platform to obtain a set of key features for immigration information, including the following steps B1-B4: B1: Based on the immigration information dataset, data cleaning techniques are used to clean the immigration information dataset and generate a cleaned immigration information dataset.

[0042] The data cleaning technique includes using a statistical outlier detection algorithm to calculate the quartiles and interquartile ranges for numerical fields such as age, income, and compensation amount, and defining outlier determination rules, expressed as: , , in, Used to define the lower boundary of the data distribution, helping to identify potential low-end outliers. Used to define the upper boundary of the data distribution, helping to identify potential high-end outliers. The interquartile range of numeric fields is used to detect and handle outliers using rules.

[0043] The detected outliers are classified and processed. Obviously erroneous data are deleted directly or marked as missing. Extreme but potentially real data are retained but marked separately. Their special characteristics are considered in subsequent analysis. An outlier processing log is recorded, including the original value, processing method, and processing time. The cleaned immigration information dataset is output.

[0044] B2: Based on the cleaned immigration information dataset, data standardization techniques are used to unify the data format, units, and coding rules to obtain a standardized immigration information dataset.

[0045] The adoption of data standardization technology to unify data formats, units, and encoding rules includes: The date format of the cleaned immigration information dataset was converted to ISO 8601 format. Extra spaces were removed, case was standardized, and special characters were deleted. Numerical values ​​of different units were unified, and categorical variables were uniformly encoded, such as gender male=1, female=2, to generate a standardized immigration information dataset.

[0046] B3: Based on the standardized immigration information dataset, use a data flow automation platform to perform data fusion on the standardized immigration information dataset to generate a fused immigration information dataset.

[0047] The process of using a data flow automation platform to perform data fusion on the standardized immigration information dataset includes: Apache NiFi (Data Stream Automation Platform) is used as the data fusion tool. Data source connectors are configured to extract data from the output buffers of four acquisition modules at regular intervals, and the same semantic fields from different data sources are mapped to a unified data pattern.

[0048] For example, income, monthly income, and household income are uniformly mapped to monthly_income (keyword).

[0049] Use immigration number as the primary key to associate records from different data sources. When the same field exists in multiple data sources, a priority rule is applied.

[0050] For example, immigration management system data > social media inferred data.

[0051] A data warehouse is built using a PostgreSQL database, and the merged data is loaded into the data warehouse for storage.

[0052] B4: Based on the merged immigration information dataset, extract key features of immigration information and construct a key feature set of immigration information.

[0053] Key features of the immigration information include: Basic immigration information characteristics: age, gender, marital status, education level, family size, and military service status.

[0054] Characteristics of immigrants' economic status: monthly income, annual income, main sources of income, family assets and liabilities.

[0055] Characteristics of immigrant resettlement information: number of months since relocation, resettlement method, total compensation amount, percentage already paid, and amount pending payment.

[0056] Immigration credit history characteristics include: number of credit cards, number of loans, number of overdue payments, longest overdue period, and third-party credit score.

[0057] Characteristics of immigrant consumer behavior: frequency of consumption, average monthly consumption amount, main consumption categories, and savings rate.

[0058] Characteristics of immigrants' social relationships: social security payment status, housing provident fund payment status, low-income status, and employment stability.

[0059] Characteristics of immigration dynamics: changes in income growth rate, employment status, and consumption levels over the past 6 months.

[0060] In step S3, based on the key feature set of immigration information, a feature importance calculation formula is constructed, high-value features are selected, and a feature importance ranking table is output, including the following steps C1-C3: C1: Based on the key feature set of immigration information, the chi-square binning algorithm is used to calculate the chi-square value of adjacent bins, and output a binning scheme corresponding to each feature.

[0061] The chi-square binning algorithm is used to calculate the chi-square value of adjacent bins, and the expression is: , in, The chi-square statistic is used to quantify the similarity between two adjacent bins in terms of the distribution of high-quality customers (no overdue payments) and low-quality customers (overdue payments) to determine whether to merge them. This represents the actual frequency. Let i be the expected frequency, i be the row index, and j be the column index.

[0062] Merge adjacent boxes with the smallest chi-square value, repeatedly calculate the chi-square value of adjacent boxes and merge adjacent boxes with the smallest chi-square value, until the stopping condition is met.

[0063] The stopping condition is: .

[0064] The final output is a binning scheme corresponding to each feature, stored in JSON format for later use.

[0065] C2: Based on the binning characteristics, construct the evidence weight calculation formula and output the evidence weight encoded dataset.

[0066] The formula for calculating the weight of evidence is expressed as follows: , in, Let be the evidence weight value of the feature of the i-th box. Let i be the number of substandard customers in the i-th box. Let be the number of high-quality customers in the i-th box. The total number of substandard customers. This represents the total number of high-quality customers.

[0067] After calculating the evidence weights of the features, an evidence weight dataset is constructed, and all feature values ​​are converted into WOE values.

[0068] C3: Based on the evidence weight encoding dataset, construct a formula for calculating feature importance, select high-value features, and output a feature importance ranking table.

[0069] In this embodiment of the application, the specific steps for constructing the feature importance calculation formula in step C3 are as follows: The formula for calculating the importance of the constructed features is expressed as follows: , , in, The quantitative representation of feature importance is a specific numerical value. This represents the contribution of the i-th bin to the overall information value (IV). Let be the evidence weight of the i-th bin, and n represent the total number of feature bins.

[0070] according to Quantify the results and set criteria for judging feature importance, including: IV < 0.02: Useless feature, delete; 0.02 ≤ IV < 0.1: Weak value feature, consider carefully; 0.1 ≤ IV < 0.3: Medium value feature, retain; 0.3 ≤ IV < 0.5: Strong value feature, use it sparingly; IV ≥ 0.5: Too high value, may be overfitting, data quality needs to be checked. Retain features with IV ∈ [0.1, 0.5] and output a feature importance ranking table.

[0071] In an alternative implementation, the feature importance calculation formula in step C3 can also be constructed using a feature importance calculation method based on the random forest algorithm. A random forest model is trained using all features and data, and the average amount of impurity reduction brought about by each feature when it is used to split nodes in all decision trees is calculated. The greater the reduction, the more important the feature. After training, the importance score of each feature is directly extracted from the trained model.

[0072] In another optional implementation, the feature importance calculation formula in step C3 can also be constructed using a recursive feature elimination method. A logistic regression model is selected and trained using all features. The weight coefficient of each feature in the model is obtained. The recursive elimination method is used to eliminate the feature with the lowest current ranking or a batch of features. The model is retrained using the remaining feature set. The performance of the new model is evaluated on the validation set. The recursive elimination is repeated to gradually reduce the number of features until the preset termination condition is reached.

[0073] It should be noted that this invention transforms the fuzzy feature selection process in traditional models into a standardized process through the WOE-IV feature engineering system. By calculating the IV value of each feature, a metric for feature prediction capability is obtained. The system can automatically complete feature selection based on a preset IV threshold, solving the problem of strong subjectivity and instability caused by traditional methods relying on expert experience or the built-in importance of the model for feature selection.

[0074] In step S4, a default risk assessment model is established and trained based on the feature importance ranking table and machine learning algorithm, and the immigrant credit score result is output, including the following steps D1-D4: D1: Establish a default risk assessment model based on the feature importance ranking table and machine learning algorithms.

[0075] D2: Train the default risk assessment model based on the key feature set of immigration information.

[0076] D3: Based on the training process of the default risk assessment model, establish a loss function to optimize the default risk assessment model.

[0077] The loss function optimization default risk assessment model is expressed as follows: , in, Represents the loss function. For the sample size, For real labels, To predict probabilities, The regularization coefficient is . denoted by , h represents the squared value of the weight of the a-th feature, h represents the total number of features, e represents the index of the current sample, and a represents the index of the current feature.

[0078] The default risk assessment model parameters are iteratively updated using the gradient descent method until the loss function reaches the maximum number of iterations, at which point the trained default risk assessment model is output.

[0079] D4: Based on the trained default risk assessment model, establish a scoring conversion formula to convert the default probability output by the default risk assessment model into a credit score.

[0080] The scoring conversion formula is expressed as follows: , , , , in, For the final credit score, Using 600 as the baseline score, derived through data experiments, the value was determined. Let the weight of the a-th feature be , Let be the WOE value of the 'a'-th feature, and 'odds' represent the proportion of good and bad customers in the current sample. This represents the baseline value for the ratio of superior to inferior customers. A represents the offset coefficient in the rating conversion formula, and B represents the scaling coefficient in the rating conversion formula. Both coefficients A and B are calculated. This represents the doubling score. Based on data experiments, a value of 20 indicates that when odds double, the credit score increases by PDO points. is the evidence weight for the baseline bin, and 'a' is the current feature index.

[0081] In this embodiment of the application, the specific steps for establishing the default risk assessment model in step S4 are as follows: The default risk assessment model is established, and its expression is: , in, For the probability of default, For the intercept term, is the total number of features, and 'a' is the current feature index. Let be the WOE value of the a-th feature.

[0082] In an optional implementation, the default risk assessment model established in step S4 can also adopt an ensemble learning model based on gradient boosting decision trees. The original features are preprocessed, such as missing value imputation, outlier handling and label encoding, the XGBoost classifier is initialized, the objective function is set to train the prediction model, and during the training process, the hyperparameters are adjusted by cross-validation to prevent overfitting. Finally, the prediction model outputs the default probability of the sample.

[0083] In another optional implementation, the default risk assessment model established in step S4 can also adopt a classification model based on support vector machines. All features are standardized. For nonlinear problems, a kernel function is selected and mapped to a high-dimensional feature space. An SVM classifier is selected and penalty parameters and kernel function parameters are set. An optimal hyperplane is found on the training set to maximize the boundary between good customer and bad customer samples. Through a probability calibration method, the output of the SVM decision function is converted into an approximate probability value output.

[0084] It should be noted that this invention employs a machine learning model based on WOE encoding. This ensures that the output score is not merely a number, but rather an interpretable decision-making process. The resulting credit score clearly demonstrates the contribution of each feature to the specific immigrant user's score, thus addressing the problem of insufficient model interpretability in the prior art. Furthermore, it can guide immigrant users on how to improve the model, enabling its long-term operation in production environments.

[0085] In step S5, after the default risk assessment model is launched, a group stability calculation formula is established to calculate the performance of the default risk assessment model and a feedback mechanism is established to update the default risk assessment model, including the following steps E1-E3: E1: After the default risk assessment model is launched, establish a group stability calculation formula to calculate the group stability of the default risk assessment model.

[0086] E2: Establish stability judgment criteria based on the group stability index of the default risk assessment model.

[0087] The stability judgment criteria include: PSI < 0.1: Model is stable, no adjustment needed; 0.1 ≤ PSI < 0.25: Model has a slight shift, attention is recommended; PSI ≥ 0.25: Model has a significant shift, retraining is required; Retraining is also triggered in the following situations: PSI ≥ 0.15 for 3 consecutive months && Actual default rate deviates from predicted default rate by more than 20% && New significant influencing factors (such as policy changes, changes in the economic environment).

[0088] E3: Based on the stability judgment criteria, establish a feedback mechanism to update the default risk assessment model.

[0089] The feedback mechanism includes: collecting new labeled data based on actual default situations, adding the new labeled data to the training set, re-performing feature engineering, model training, and score conversion, evaluating the performance of the new model on the validation set, and deploying the new model if the performance is improved, otherwise retaining the original model.

[0090] In this embodiment of the application, the specific steps for establishing the population stability calculation formula in step S5 are as follows: The formula for calculating group stability is expressed as follows: , in, As a population stability index, The total number of rating intervals. Let be the index of the pi-th rating interval. This represents the proportion of the actual sample in the pi-th scoring interval. This represents the proportion of the expected sample in the pi-th scoring interval.

[0091] In an optional implementation, the population stability calculation formula in step S5 can also be established using a monitoring method based on model performance indicators. When the model is launched, its core performance indicators on the test set are recorded as benchmark values. Every month or quarter, a batch of samples that have recently produced actual results are extracted from the production system. The current model is used to predict these review samples, calculate the same performance indicators, obtain the current values, and compare the current values ​​with the benchmark values ​​to obtain the performance evaluation results.

[0092] In another optional implementation, the population stability calculation formula in step S5 can also be established using a feature stability monitoring method. Based on the feature distribution of the model training period data, the difference between the current production data and the benchmark data in terms of statistics, such as the relative rate of change of the mean, standard deviation, and median, is calculated. The change in the proportion of each category is calculated, and an alarm threshold for the distribution change of each important feature is set and an alarm is triggered.

[0093] It should be noted that the group stability index monitoring used in this invention can directly pinpoint the location of problems. By directly comparing the distribution changes of model output scores over time, it can detect the decline in the model's matching degree with the current customer group before the model's actual business performance undergoes an observable decline, providing advance notice for model retraining or optimization and reducing the financial risks caused by model failure.

[0094] In summary, this invention constructs an immigrant credit risk assessment system based on multi-source data fusion, WOE-IV feature engineering, interpretable machine learning modeling, and dynamic feedback optimization. This system enables fully automated management of the immigrant population from data collection to feature quantification to credit score prediction, solving the problems of single data dimensions, poor model interpretability, and lack of dynamic tracking mechanisms in traditional credit assessment methods.

[0095] Example 3 illustrates a schematic scheme of an immigration management method based on machine learning. It should be noted that the technical solution of this machine learning-based immigration management system and the technical solution of the aforementioned machine learning-based immigration management method belong to the same concept. Details not described in detail in this example of the machine learning-based immigration management system can be found in the description of the aforementioned machine learning-based immigration management method.

[0096] This embodiment also provides a machine learning-based immigration management system, including a data acquisition module, a preprocessing module, a feature calculation module, an output module, and a feedback module; The acquisition module is used to establish a multi-source data acquisition system to collect immigration information data and construct an immigration information dataset. The preprocessing module is used to preprocess the immigration information dataset using a data flow automation platform to obtain a set of key features of immigration information. The feature calculation module is used to construct a feature importance calculation formula based on the key feature set of immigration information, screen out high-value features, and output a feature importance ranking table. The output module is used to establish and train a default risk assessment model based on a feature importance ranking table and a machine learning algorithm, and output the immigrant credit score result. The feedback module is used to calculate the performance of the default risk assessment model by establishing a group stability calculation formula after the default risk assessment model is launched, and to establish a feedback mechanism to update the default risk assessment model.

[0097] This embodiment also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of a machine learning-based immigration management method.

[0098] This embodiment proposes a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the steps of a machine learning-based immigration management method.

[0099] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0100] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A machine learning-based immigration management method, characterized in that, include: Establish a multi-source data collection system to collect immigration information data and construct an immigration information dataset; Based on the immigration information dataset, an automated data flow platform was used to preprocess the immigration information dataset to obtain a set of key features of immigration information. Based on the key feature set of immigration information, a feature importance calculation formula is constructed, high-value features are screened out, and a feature importance ranking table is output. Based on the feature importance ranking table and machine learning algorithm, a default risk assessment model is established and trained to output immigrant credit score results. After the default risk assessment model is launched, a group stability calculation formula is established to calculate the performance of the default risk assessment model, and a feedback mechanism is established to update the default risk assessment model.

2. The immigration management method based on machine learning as described in claim 1, characterized in that, The establishment of a multi-source data collection system to collect immigration information data and construct an immigration information dataset includes: Database connection pooling technology is used to collect basic immigration information; We use API interfaces to connect with bank credit reporting centers and third-party credit reporting platforms to collect immigration credit data. Web crawling technology was used to collect data on the social behavior characteristics of immigrants; Collect immigrant socioeconomic data using a government data exchange platform; Establish a data integrity monitoring and feedback mechanism to perform integrity checks on the collected data, and trigger the feedback mechanism when a data integrity problem is detected; An immigrant information dataset is constructed based on the immigrant basic information, immigrant credit data, immigrant social behavior characteristics data, and immigrant socioeconomic data that have passed integrity testing.

3. The immigration management method based on machine learning as described in claim 2, characterized in that, The process involves preprocessing the immigration information dataset using a data flow automation platform to obtain a set of key features for immigration information, including: Based on the immigration information dataset, data cleaning techniques are used to clean the immigration information dataset and generate a cleaned immigration information dataset. Based on the cleaned immigration information dataset, data standardization techniques are used to unify the data format, units, and coding rules to obtain a standardized immigration information dataset. Based on the standardized immigration information dataset, a data flow automation platform is used to perform data fusion on the standardized immigration information dataset to generate a fused immigration information dataset. Based on the merged immigration information dataset, an information value calculation formula is established, key features of immigration information are extracted, and a key feature set of immigration information is constructed.

4. The immigration management method based on machine learning as described in claim 3, characterized in that, Based on the key feature set of immigration information, a feature importance calculation formula is constructed to screen out high-value features and output the feature importance. The ranking table includes: Based on the key feature set of immigration information, the chi-square binning algorithm is used to calculate the chi-square value of adjacent bins, and output a binning scheme corresponding to each feature. Based on the binning characteristics, construct the evidence weight calculation formula and output the evidence weight encoded dataset; Based on the evidence weight encoding dataset, a formula for calculating feature importance is constructed, high-value features are selected, and a feature importance ranking table is output.

5. The immigration management method based on machine learning as described in claim 4, characterized in that, The default risk assessment model is established and trained based on a feature importance ranking table combined with machine learning algorithms, and outputs an immigrant credit score, including: Based on a feature importance ranking table and machine learning algorithms, a default risk assessment model is established, with the following expression: ， in, For the probability of default, For the intercept term, is the total number of features, and 'a' is the current feature index. Let be the WOE value of the a-th feature; The default risk assessment model is trained based on the key feature set of immigration information. Based on the training process of the default risk assessment model, a loss function is established to optimize the default risk assessment model, and the expression is: ， in, Represents the loss function. For the sample size, For real labels, To predict probabilities, The regularization coefficient is . represents the squared value of the weight of the a-th feature, h represents the total number of features, e represents the index of the current sample, and a represents the index of the current feature; The default risk assessment model parameters are iteratively updated using the gradient descent method until the loss function reaches the maximum number of iterations, and the trained default risk assessment model is then output. Based on the trained default risk assessment model, a scoring conversion formula is established to convert the default probability output by the default risk assessment model into a credit score. The expression is as follows: ， ， ， ， in, For the final credit score, The baseline score was obtained through data experiments. Let the weight of the a-th feature be , Let be the WOE value of the 'a'-th feature, and 'odds' represent the proportion of good and bad customers in the current sample. This represents the baseline value for the ratio of superior to inferior customers; A represents the offset coefficient in the rating conversion formula; and B represents the scaling coefficient in the rating conversion formula. This indicates a doubling of the score, derived from data experiments. It means that when the number of odds doubles, the credit score increases by the PDO score. is the evidence weight for the baseline bin, and 'a' is the current feature index.

6. The immigration management method based on machine learning as described in claim 5, characterized in that, After the default risk assessment model is launched, a group stability calculation formula is established to calculate the performance of the default risk assessment model, and a feedback mechanism is established to update the default risk assessment model, including: After the default risk assessment model is launched, a group stability calculation formula is established to calculate the group stability of the default risk assessment model. The expression is as follows: ， in, As a population stability index, The total number of rating intervals. Let be the index of the pi-th rating interval. This represents the proportion of the actual sample in the pi-th scoring interval. This represents the proportion of the expected sample in the pi-th scoring interval; Based on the group stability index of the default risk assessment model, establish stability judgment criteria; Based on stability assessment criteria, a feedback mechanism is established to update the default risk assessment model.

7. The immigration management method based on machine learning as described in claim 6, characterized in that, The process involves encoding the dataset based on evidence weights, constructing a formula for calculating feature importance, selecting high-value features, and outputting the feature importance. The ranking table includes: The formula for calculating the importance of the features is expressed as follows: ， ， in, The quantitative representation of feature importance is a specific numerical value. This represents the contribution of the i-th bin to the overall information value (IV). Let n be the evidence weight for the i-th bin, and n represent the total number of feature bins. Let i be the number of substandard customers in the i-th box. Let be the number of high-quality customers in the i-th box. The total number of substandard customers. This represents the total number of high-quality customers.

8. A machine learning-based immigration management system, employing the method described in any one of claims 1-7, characterized in that, It includes an acquisition module, a preprocessing module, a feature calculation module, an output module, and a feedback module; The acquisition module is used to establish a multi-source data acquisition system to collect immigration information data and construct an immigration information dataset. The preprocessing module is used to preprocess the immigration information dataset using a data flow automation platform to obtain a set of key features of immigration information. The feature calculation module is used to construct a feature importance calculation formula based on the key feature set of immigration information, screen out high-value features, and output a feature importance ranking table. The output module is used to establish and train a default risk assessment model based on a feature importance ranking table and a machine learning algorithm, and output the immigrant credit score result. The feedback module is used to calculate the performance of the default risk assessment model by establishing a group stability calculation formula after the default risk assessment model is launched, and to establish a feedback mechanism to update the default risk assessment model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the immigration management method based on machine learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the machine learning-based immigration management method according to any one of claims 1 to 7.

Images