Real estate financial risk prevention and control method and system based on artificial intelligence

Abstract

The invention provides a real estate financial risk prevention and control method and system based on artificial intelligence, and the method comprises the steps: collecting structured data and unstructured data of real estate registration information, and carrying out the feature extraction of the structured data and the unstructured data; constructing a multi-dimensional scoring system, and calculating a comprehensive score of each real estate according to the quantitative score and the weight of each index; selecting a model architecture, constructing a risk prediction model based on a neural network architecture, and training and optimizing the risk prediction model by taking the extracted feature data and the multi-dimensional comprehensive score as input; calculating the risk score of the real estate through the trained model; and determining a risk level according to the risk score and a preset threshold value, and when the risk level exceeds the preset threshold value, giving a risk early warning. The real estate financial risk can be effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the field of financial risk prevention and control, and specifically relates to a real estate financial risk prevention and control method and system based on artificial intelligence. Background Art

[0002] In the field of real estate finance, risk prevention and control is a key link in ensuring financial stability and asset security. Traditional real estate financial risk prevention and control methods mainly rely on manual experience and simple statistical analysis, which has many limitations.

[0003] On the one hand, data processing efficiency is low. Faced with massive amounts of real estate registration information, macroeconomic data, and market dynamics data, manual processing methods are difficult to quickly and accurately screen and analyze key information, resulting in a long risk assessment cycle and an inability to respond to market changes in a timely manner.

[0004] On the other hand, the accuracy of risk assessment is insufficient. Simple statistical analysis methods are difficult to capture complex and changing market factors and potential risks, and are prone to overlooking some hidden risk signals, which makes financial institutions face higher risk exposure in real estate finance business.

[0005] In addition, traditional methods lack effective use of unstructured data. Unstructured data such as real estate-related contract texts, policy and regulatory documents contain rich risk information, but traditional methods find it difficult to extract valuable features from them, resulting in incomplete and in-depth risk assessment.

[0006] With the rapid development of artificial intelligence technology, it has shown great advantages in data processing, pattern recognition and predictive analysis. However, in the field of real estate financial risk prevention and control, the application of artificial intelligence technology is not mature and perfect enough, and there is a lack of a systematic, comprehensive and efficient artificial intelligence-based prevention and control system. Summary of the invention

[0007] The purpose of the present invention is to provide a real estate financial risk prevention and control method and system based on artificial intelligence to improve the accuracy and efficiency of real estate financial risk assessment, timely discover potential risks and take effective prevention and control measures, reduce the risk losses of financial institutions in real estate financial business, and ensure the stable operation of the financial market.

[0008] The technical solution of the present invention is as follows:

[0009] A real estate financial risk prevention and control method based on artificial intelligence, comprising the following steps:

[0010] Collecting structured data and unstructured data of real estate registration information, and performing feature extraction on the structured data and unstructured data;

[0011] Construct a multi-dimensional scoring system to calculate the comprehensive score of each real estate based on the quantitative scores and weights of each indicator;

[0012] Select a model architecture, build a risk prediction model based on a neural network architecture, use the extracted feature data and multi-dimensional comprehensive scores as input, and train and optimize the risk prediction model;

[0013] Calculate the risk score of real estate through the trained model;

[0014] The risk level is determined based on the risk score and preset thresholds. When the risk level exceeds the set threshold, a risk warning is issued.

[0015] Furthermore, the structured data and unstructured data collected from real estate registration information specifically include:

[0016] Obtain basic real estate information, property rights information, and transaction record structured data from the real estate registration management department;

[0017] Collect unstructured data of real estate-related contract texts, appraisal reports, and policy and regulatory documents;

[0018] At the same time, external data such as macroeconomic data, market supply and demand data, and interest rate data are collected as auxiliary information for risk assessment.

[0019] Furthermore, feature extraction is performed on the structured data and the unstructured data, specifically:

[0020] For structured data, numerical features related to risk assessment were screened, and features with a significance threshold less than 0.05 were selected as valid features through a one-way ANOVA test;

[0021] For unstructured data, perform word segmentation, cleaning and normalization, calculate the TF-IDF value and build a feature matrix.

[0022] Furthermore, the construction of a multi-dimensional scoring system includes:

[0023] Determine the scoring dimensions, including the real estate dimension, market environment dimension and financial status dimension;

[0024] Quantify the specific indicators under each dimension and use the hierarchical analysis method or Delphi method to assign corresponding weights to each dimension and indicator;

[0025] Based on the quantitative scores of each indicator and the corresponding weights, a comprehensive score for each property is calculated.

[0026] Furthermore, the dimensions of the movable property itself include: geographical location, building quality, and supporting facilities; the dimensions of the market environment include: macroeconomic data, market supply and demand, and industry development trends; and the dimensions of the financial status include: credit status, debt level, and repayment ability.

[0027] Furthermore, the risk prediction model is trained and optimized, specifically including:

[0028] Divide the data set into training set, validation set and test set according to a certain ratio;

[0029] Initialize the weights and bias parameters of the neural network;

[0030] The risk prediction model is trained using the training set, the prediction output is calculated by forward propagation, and a loss function is defined to measure the difference between the prediction output and the true label;

[0031] Use the gradient descent algorithm for back propagation to update the model parameters;

[0032] Use the validation set to adjust the model’s hyperparameters;

[0033] Finally, the test set is used to evaluate the final performance of the model.

[0034] Furthermore, the risk level is determined according to the risk score and the preset threshold, specifically:

[0035] Set the threshold for risk classification and divide the risk level into three levels: low, medium and high;

[0036] Calculate the quantile of the risk score output by the risk prediction model, map the calculated quantile to the risk level, and obtain a determined risk level.

[0037] The present invention also provides a real estate financial risk prevention and control system based on artificial intelligence, comprising:

[0038] A data collection module, used to collect structured data and unstructured data of real estate registration information;

[0039] A feature extraction module, used for extracting features from the structured data and the unstructured data;

[0040] A multi-dimensional scoring system construction module is used to construct a multi-dimensional scoring system and calculate the comprehensive score of each real estate based on the quantitative scores and weights of each indicator;

[0041] A model building and training module is used to select a model architecture, build a risk prediction model based on a neural network architecture, and train and optimize the risk prediction model using the extracted feature data and multi-dimensional comprehensive scores as input;

[0042] The risk assessment module is used to calculate the risk score of real estate through the trained model;

[0043] The risk warning module is used to determine the risk level based on the risk score and the preset threshold, and issue a risk warning when the risk level exceeds the set threshold.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] The present invention collects structured and unstructured data of real estate registration and various external data, constructs a multi-dimensional scoring system, comprehensively considers indicators of dimensions such as real estate itself, market environment and financial status, quantifies indicators and assigns weights, and ensures that risk assessment is more accurate and comprehensive; builds a risk prediction model based on a multi-layer perceptron architecture, and at the same time, establishes a real-time data collection and processing system, which can conduct risk assessment and early warning in a timely manner, thereby buying precious time for financial institutions and relevant departments to respond to risks, effectively reducing real estate financial risks, and ensuring the stable development of the industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the description and claims, serve to illustrate the embodiments of the invention. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0047] Figure 1 A schematic flow chart of the method of the present invention is shown. DETAILED DESCRIPTION

[0048] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0049] like Figure 1 As shown, the present invention provides a real estate financial risk prevention and control method based on artificial intelligence;

[0050] 1. Data collection and feature extraction

[0051] Data Collection

[0052] Acquisition of structured data: The latest basic real estate information is obtained through the real estate registration management department, including the house address, area, and apartment structure; property information, including the name of the owner, ID number, and property share; and transaction records for the past year, covering structured data such as transaction time, transaction price, and information about both parties to the transaction. These data are stored in the relational database MySQL to ensure structured storage and efficient query of data.

[0053] Unstructured data collection: Using web crawler technology, unstructured data such as contract texts, appraisal reports, policy and regulatory documents related to real estate are obtained and extracted from government official websites, well-known real estate information platforms, and industry authoritative forums, and stored according to different sources and types for subsequent centralized processing and analysis.

[0054] External data integration: Macroeconomic data such as quarterly GDP growth rate and annual inflation rate, market supply and demand data such as monthly new house supply and second-hand house transaction volume, and interest rate data including central bank benchmark interest rate and commercial bank mortgage interest rate are obtained through government statistical departments, financial data providers, and professional market research institutions. These data are stored in the data warehouse in a standardized format for convenient correlation analysis with real estate-related data.

[0055] Feature extraction

[0056] Structured data feature extraction: Conduct a one-way ANOVA test on structured data. Before the test, identify and process outliers using the IQR method, and then use the Min-Max normalization method to map the data to the [0,1] interval. During the ANOVA test, set the significance level to 0.05, test each numerical feature against the risk level, and select features with a p-value less than 0.05, such as the volatility coefficient of property prices and the difference between the property age and the average property age of surrounding properties, as effective structured features for subsequent risk assessment.

[0057] Unstructured data feature extraction: First, perform word segmentation to split the text into individual words or phrases; then, use a stop word list to remove common meaningless words such as "of", "is", "in", etc.; next, through stemming or lemmatization techniques, unify the words to their basic form, for example, reducing "running" and "runs" to "run". On this basis, use the TfidfVectorizer class in the scikit-learn library to calculate TF-IDF values and construct a feature matrix to convert the text data into numerical features for subsequent machine learning model processing.

[0058] (2) Construction of a multi-dimensional scoring system

[0059] Determine the scoring dimensions and indicators

[0060] Real estate dimensions: A team of senior real estate appraisers and urban planning experts will determine geographical location indicators based on urban development plans and real estate market research data, and score properties based on different regional divisions such as urban core areas, secondary core areas, and suburbs, as well as factors such as the distance to surrounding transportation hubs (such as subway stations and bus stations) and the proximity to commercial centers. Building quality indicators are evaluated based on the age of construction, building structure type (such as frame structure, brick-concrete structure), quality grade of building materials, and maintenance records of the property. Supporting facilities indicators take into account the number and quality of surrounding public facilities such as schools, hospitals, and parks, and quantify scores based on the school's education quality ranking, hospital grade, park area, and degree of facility completeness.

[0061] Market environment dimension: Macroeconomic analysts and real estate market research experts determine macroeconomic data indicators based on the fluctuation trend of macroeconomic data, changes in market supply and demand, and the latest developments in the industry, such as the year-on-year change in GDP growth rate and the comparison of inflation rate with the industry average; market supply and demand relationship indicators are evaluated by calculating the housing inventory turnover cycle, the ratio of new housing supply to demand, etc.; industry development trend indicators focus on the adjustment direction of real estate policies, the application degree of emerging technologies (such as smart homes and green building technologies) in the industry, etc. for scoring.

[0062] Financial status dimension: Financial risk control experts and credit assessment agency personnel determine credit status indicators based on the credit data of financial institutions, credit rating reports and industry-standard financial indicator standards, and score with reference to the scores of credit rating agencies and the credit history of individuals or enterprises (such as the number of overdue repayments and default records); debt level indicators are evaluated by calculating financial indicators such as the debt-to-asset ratio and debt-to-income ratio; repayment ability indicators are quantitatively scored based on income stability (such as the diversity of income sources, fluctuations in income in the past three years) and cash flow status (such as monthly net cash flow, the number of times cash flow covers debt).

[0063] Indicator quantification and weight allocation

[0064] Analytic Hierarchy Process (AHP): Construct a judgment matrix and compare the indicators in each dimension in pairs. For example, in the dimension of real estate itself, compare the relative importance of the three indicators of geographical location, building quality, and supporting facilities. Use a 1-9 scale, where 1 means that the two indicators are equally important, 3 means that one indicator is slightly more important than the other, 5 means obviously important, 7 means strongly important, 9 means extremely important, and 2, 4, 6, and 8 are intermediate values. Determine the value of the judgment matrix through multiple expert discussions and feedback. Then, calculate the maximum eigenvalue and eigenvector of the judgment matrix, normalize the eigenvector, and obtain the weight of each indicator.

[0065] Delphi method: Design a detailed expert questionnaire and invite 10-15 experts in the fields of finance, real estate, and economics to participate. The questionnaire lists the indicators under each dimension and asks the experts to score the importance of the indicators on a scale of 1-10, with 1 being the least important and 10 being the most important. After collecting the feedback from the experts, conduct statistical analysis and calculate the mean and standard deviation of the scores for each indicator. For indicators with large differences in scores, conduct a second round of questionnaires and feedback the statistical results of the first round to let the experts reconsider and adjust the scores. After 2-3 rounds of questionnaires and feedback, the weights of each indicator are finally determined.

[0066] Indicator quantification: For qualitative indicators, detailed quantitative standards are formulated. For example, for geographical location indicators, the core area of ​​the city is assigned 8-10 points, the secondary core area is assigned 5-7 points, and the suburbs are assigned 1-4 points; for building quality indicators, newly built buildings that use high-quality building materials and are well maintained are assigned 8-10 points, older buildings with stable structures and general maintenance are assigned 4-7 points, and buildings with obvious quality problems are assigned 1-3 points. For quantitative indicators, different intervals are divided and assigned according to the distribution of data. For example, an asset-liability ratio below 30% is assigned 8-10 points, 30%-60% is assigned 4-7 points, and above 60% is assigned 1-3 points.

[0067] Comprehensive score calculation: Based on the quantitative scores of each indicator and the corresponding weights, the weighted summation method is used to calculate the comprehensive score of each real estate. For example, the geographical location score of a real estate in the real estate dimension is 8 points, with a weight of 0.4; the building quality score is 7 points, with a weight of 0.3; the supporting facilities score is 6 points, with a weight of 0.3. Then the score of this dimension is 8×0.4+7×0.3+6×0.3=7.1 points. Similarly, the scores of the market environment dimension and the financial status dimension are calculated, and finally the comprehensive score of the real estate is calculated based on the weights of the three dimensions.

[0068] 3. Risk prediction model construction and training

[0069] Model selection and architecture construction: Based on the analysis of data features and the need for risk assessment in the early stage, the multi-layer perceptron (MLP) architecture is selected. The model is built using the TensorFlow framework. First, the number of neurons in the input layer is determined to be consistent with the dimension of the extracted feature data. For example, if 100 valid features are obtained after feature extraction, the number of neurons in the input layer is set to 100. The number of hidden layers is determined based on experiments and experience, and is generally set to 2-3 layers for trial. The number of neurons in each hidden layer is set in a gradually decreasing manner, such as 128 neurons in the first hidden layer, 64 neurons in the second layer, and 32 neurons in the third layer. A single neuron is set in the output layer to output the risk score. The layers are connected in a fully connected manner, that is, each neuron in the previous layer is connected to each neuron in the next layer, and neurons in adjacent layers are connected through weight matrices.

[0070] Model training and optimization:

[0071] Data division: The data set is divided into 70% training set, 20% validation set, and 10% test set.

[0072] Parameter initialization: The Xavier initialization method is used to initialize the weights and bias parameters of the neural network.

[0073] Model training: Define the loss function as the cross entropy loss function (for classification problems) or the mean square error loss function (for regression problems, where risk scoring can be considered a regression problem), use gradient descent algorithms such as Adam for back propagation, and update model parameters. Specifically, in TensorFlow, define optimizer = tf.keras.optimizers.Adam (learning_rate = 0.001) to set the optimizer and learning rate. During the training process, set the number of training rounds (such as 100 rounds), and in each round of training, input the data in the training set into the model in batches according to the batch size (such as 32) for training.

[0074] Hyperparameter adjustment: Use the validation set to adjust the model's hyperparameters. For example, the learning rate is adjusted in the range of 0.001-0.1. Through multiple experiments, the performance indicators of the model on the validation set (such as root mean square error, accuracy, etc.) are observed, and the learning rate with the best performance is selected. The batch size is adjusted between 16-128, and the optimal batch size is also determined based on the performance of the validation set. The number of neurons in the hidden layer can also be increased or decreased according to the experimental results, such as increasing or decreasing the number of hidden layers, and adjusting the number of neurons in each hidden layer to find the most suitable architecture for the model.

[0075] Model evaluation: Use the test set to perform a final evaluation on the optimized model and calculate the model's accuracy, recall, F1 value, root mean square error and other evaluation indicators. Use the evaluation results as the final measure of model performance to determine whether the model meets the needs of actual applications.

[0076] (IV) Risk assessment and early warning

[0077] Risk assessment: Establish a real-time data collection and processing system to obtain the latest real estate-related data in real time through the data interface, including market transaction data, policy and regulatory updates, changes in macroeconomic indicators, etc. The real-time collected data is preprocessed, such as data cleaning, feature extraction, normalization, etc., to make it meet the input requirements of the trained model. Then, the preprocessed data is input into the trained risk prediction model. The model calculates the risk score of the real estate based on the input feature data and multi-dimensional comprehensive scores. For example, the risk score output by the model is a value between 0-10, and the larger the value, the higher the risk.

[0078] Risk warning: Set a risk level threshold, such as classifying risk scores of 0-3 as low risk, 3-7 as medium risk, and 7-10 as high risk. When the risk level corresponding to the risk score calculated by the model exceeds the set threshold (such as high risk), a warning message is issued through the SMS interface (such as using Alibaba Cloud SMS service), mail server (such as using NetEase Enterprise Mailbox's SMTP service) or system pop-up window (setting pop-up window reminders on the user interface of the risk prevention and control system). The warning information contains the basic information of the real estate, risk score, risk level, and analysis of possible causes of the risk, reminding the risk management department, business decision-makers and other relevant personnel of the financial institution to pay attention and take corresponding risk prevention and control measures in a timely manner, such as adjusting credit policies, increasing collateral requirements, and strengthening post-loan supervision.

[0079] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A real estate financial risk prevention and control method based on artificial intelligence, characterized in that: The following steps are involved: Collecting structured data and unstructured data of real estate registration information, and performing feature extraction on the structured data and unstructured data; Construct a multi-dimensional scoring system to calculate the comprehensive score of each real estate based on the quantitative scores and weights of each indicator; Select a model architecture, build a risk prediction model based on a neural network architecture, use the extracted feature data and multi-dimensional comprehensive scores as input, and train and optimize the risk prediction model; Calculate the risk score of real estate through the trained model; The risk level is determined based on the risk score and preset thresholds. When the risk level exceeds the set threshold, a risk warning is issued.

2. The method for preventing and controlling real estate financial risks based on artificial intelligence according to claim 1 is characterized in that: The structured data and unstructured data collected from real estate registration information specifically include: Obtain basic real estate information, property rights information, and transaction record structured data from the real estate registration management department; Collect unstructured data of real estate-related contract texts, appraisal reports, and policy and regulatory documents; At the same time, external data such as macroeconomic data, market supply and demand data, and interest rate data are collected as auxiliary information for risk assessment.

3. The method for preventing and controlling real estate financial risks based on artificial intelligence according to claim 1 is characterized in that: Feature extraction is performed on the structured data and the unstructured data, specifically: For structured data, numerical features related to risk assessment were screened, and features with a significance threshold less than 0.05 were selected as valid features through a one-way ANOVA test; For unstructured data, perform word segmentation, cleaning and normalization, calculate the TF-IDF value and build a feature matrix.

4. The method for preventing and controlling real estate financial risks based on artificial intelligence according to claim 1 is characterized in that: The multi-dimensional scoring system is constructed, including: Determine the scoring dimensions, including the real estate dimension, market environment dimension and financial status dimension; Quantify the specific indicators under each dimension and use the hierarchical analysis method or Delphi method to assign corresponding weights to each dimension and indicator; Based on the quantitative scores of each indicator and the corresponding weights, a comprehensive score for each property is calculated.

5. The method for preventing and controlling real estate financial risks based on artificial intelligence according to claim 1 is characterized in that: The dimensions of the movable property itself include: geographical location, construction quality, and supporting facilities; the dimensions of the market environment include: macroeconomic data, market supply and demand, and industry development trends; the dimensions of the financial status include: credit status, debt level, and repayment ability.

6. The method for preventing and controlling real estate financial risks based on artificial intelligence according to claim 1, characterized in that: The risk prediction model is trained and optimized, specifically including: Divide the data set into training set, validation set and test set according to a certain ratio; Initialize the weights and bias parameters of the neural network; The risk prediction model is trained using the training set, the prediction output is calculated by forward propagation, and a loss function is defined to measure the difference between the prediction output and the true label; Use the gradient descent algorithm for back propagation to update the model parameters; Use the validation set to adjust the model’s hyperparameters; Finally, the test set is used to evaluate the final performance of the model.

7. The method for preventing and controlling real estate financial risks based on artificial intelligence according to claim 1 is characterized in that: The risk level is determined based on the risk score and the preset threshold, specifically: Set the threshold for dividing risk levels into three levels: low, medium and high; Calculate the quantile of the risk score output by the risk prediction model, map the calculated quantile to the risk level, and obtain a determined risk level.

8. An artificial intelligence-based real estate financial risk prevention and control system, characterized in that: include: A data collection module, used to collect structured data and unstructured data of real estate registration information; A feature extraction module, used for extracting features from the structured data and the unstructured data; A multi-dimensional scoring system construction module is used to construct a multi-dimensional scoring system and calculate the comprehensive score of each real estate based on the quantitative scores and weights of each indicator; A model building and training module is used to select a model architecture, build a risk prediction model based on a neural network architecture, and train and optimize the risk prediction model using the extracted feature data and multi-dimensional comprehensive scores as input; The risk assessment module is used to calculate the risk score of real estate through the trained model; The risk warning module is used to determine the risk level based on the risk score and the preset threshold, and issue a risk warning when the risk level exceeds the set threshold.

Images