Real estate value prediction method, electronic equipment, computer readable medium and program product

Abstract

The invention provides a real estate value prediction method, electronic equipment, a computer readable medium and a program product. The real estate value prediction method comprises the following steps: S1, collecting data information related to real estate to establish a first data set, wherein the first data set comprises a plurality of independent variables and at least one dependent variable; s2, calculating respective feature importance scores of the plurality of independent variables based on a first random forest model, and deleting the independent variables of which the feature importance scores are smaller than a threshold value in the first data set so as to obtain a second data set; step S3, five basic models are constructed, and the five basic models are fused by using a Stacking algorithm, so that an optimal meta-learner is selected to construct a Stacking fusion model; and S4, optimizing the Stacking fusion model by using a precision weighting strategy so as to obtain a real estate value prediction model.

Description

Technical Field

[0001] The present invention relates to the fields of integrated learning and automatic machine learning, and in particular to a method for predicting the value of real estate, an electronic device, a computer-readable medium and a program product. Background Art

[0002] In the construction of non-standard factor market, real estate value prediction is an important link in the leasing and disposal process of state-owned commercial real estate. Its accuracy and efficiency directly affect the efficient circulation of factor resources and the improvement of market mechanism. However, traditional real estate value prediction methods, such as cost method, income method, market method, etc., have long prediction cycles, high costs and are easily affected by subjective factors, and are difficult to meet the market demand for efficient circulation of factor resources.

[0003] By using digital means and technical algorithm models, combined with the advantages of basic models and Stacking fusion models, we explore more efficient and accurate real estate value prediction methods. This will not only help reduce the cycle and cost of state-owned commercial real estate transactions, but also accelerate the circulation of factor resources, improve transaction efficiency, and thus create a fair, open and just state-owned commercial real estate transaction mechanism and improve operational efficiency.

[0004] As an advanced integrated learning technology, the Stacking fusion model has demonstrated powerful predictive capabilities in many fields. This technology has been widely used in complex scenarios such as passenger flow prediction, ship fuel consumption prediction, and carbon emission prediction, and can effectively solve the problem of insufficient predictive capabilities of a single model. However, most of these applications focus on the data characteristics of specific industries. For the field of state-owned commercial real estate, which has unique market attributes and complex influencing factors, existing research is still insufficient.

[0005] More importantly, the existing stacking fusion models are mostly traditional architectures that fail to fully consider the performance differences between base learners and their contribution to the final prediction results. In practical applications, the prediction accuracy of different base learners often varies due to differences in algorithm principles, data processing capabilities and other factors. In view of this, there is an urgent need for a more accurate and efficient real estate value prediction method that can target the characteristics of state-owned commercial real estate value prediction. Summary of the invention

[0006] In order to solve at least some of the above technical problems, the present application provides a method for predicting the value of real estate, an electronic device, a computer-readable medium and a program product, which can accurately and efficiently predict the value of real estate in the non-standard factor market.

[0007] The first aspect of the present application provides a method for predicting the value of real estate, comprising the following steps: step S1, obtaining data information related to real estate to establish a first data set, and preprocessing the first data set, wherein the first data set includes multiple independent variables and at least one dependent variable, and the data information related to real estate includes at least one of the following information: project information, location information, asset information, transaction description information, or matter description information; step S2, calculating the feature importance scores of the multiple independent variables based on a first random forest model, setting a threshold according to the distribution of the feature importance scores, and deleting the independent variables in the first data set whose feature importance scores are less than the threshold, thereby obtaining a second data set; step S3, constructing a BP neural network model, a second random forest model, and an XGBoost model , LightGBM model and CatBoost model, and use the TPOT automated machine learning tool to optimize the hyperparameters of the five basic models respectively, select one of the basic models as the meta-learner in turn, and the remaining four basic models as base learners, use the second data set and use the Stacking algorithm to fuse the five basic models respectively, so as to construct five Stacking fusion models, and select the optimal Stacking fusion model from the five Stacking fusion models through model performance evaluation, and the meta-learner corresponding to the optimal Stacking fusion model is the optimal meta-learner; step S4, use the precision weighted strategy to optimize the optimal Stacking fusion model, so as to obtain a real estate value prediction model.

[0008] In the first aspect of the present application, step S1 includes: step S11, without changing the distribution characteristics of the first data set, using masking technology to desensitize the first data set; step S12, using the K nearest neighbor algorithm to fill in the data of the first data set; step S13, the multiple independent variables include categorical independent variables and numerical independent variables, using one-hot encoding to convert each of the categorical independent variables into a set of binary vectors, and using range normalization to scale each of the numerical independent variables to the interval [0,1].

[0009] In the first aspect of the present application, step S13 further includes: the at least one dependent variable is the real estate value, and a logarithmic transformation is performed on the dependent variable to reduce the skewed distribution.

[0010] In the first aspect of the present application, step S2 includes: step S21, using a first random forest model to model the multiple independent variables and the at least one dependent variable, the first random forest model includes multiple decision trees, each decision tree randomly selects at least one independent variable from the multiple independent variables for node splitting during training, so as to predict the importance of each independent variable; step S22, for each decision tree, calculate the Gini impurity reduction or information gain of each independent variable at all split nodes in the decision tree, and then average the calculation results of all decision trees to obtain the feature importance score of each independent variable; step S23, according to the calculated feature importance scores, sort the multiple independent variables in descending order, and set a threshold according to the distribution of the feature importance scores, delete the independent variables in the first data set whose feature importance scores are less than the threshold, so as to obtain the second data set.

[0011] In the first aspect of the present application, step S4 further includes: step S41, using the second data set to perform five-fold cross validation, for each base learner i=1,2,3,4, based on the prediction result in the j-th fold cross validation and the corresponding true value Calculate the mean square error:

[0012]

[0013] In the formula, m valid is the number of samples in the validation subset,

[0014] Based on the mean square error, calculate the accuracy of each base learner in the j-fold cross validation:

[0015] Accuracy i，j =1-MSE i，j

[0016] For each base learner, calculate its average accuracy in the five-fold cross validation:

[0017]

[0018] Step S42: In each training iteration, based on the average accuracy, determine the dynamic weight ω i :

[0019]

[0020] Step S43: In the test phase, for each test sample, four base learners are used to perform predictions respectively to obtain prediction results. Based on the dynamic weight ω i , perform weighted averaging on the prediction results to obtain a weighted average prediction result:

[0021]

[0022] Step S44: using the weighted average prediction result as new test data for testing the optimal meta-learner, optimizing the optimal Stacking fusion model, and thereby obtaining a real estate value prediction model.

[0023] The second aspect of the present application provides an electronic device, comprising: one or more processors; one or more memories; the one or more memories storing one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device executes the real estate value prediction method described in any one of claims 1 to 5.

[0024] A third aspect of the present application provides a computer-readable medium having instructions stored thereon, which, when executed on a computer, causes the computer to execute the method for predicting the value of real estate according to any one of claims 1 to 5.

[0025] A fourth aspect of the present application provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the method for predicting the value of real estate described in any one of claims 1 to 5.

[0026] The real estate value prediction method provided in this application constructs a real estate value assessment index system, builds a Stacking fusion model for the indicator variables after feature selection, and performs precision weighted optimization, thereby achieving accurate and efficient prediction of real estate value. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flowchart of a method according to an embodiment of the present application;

[0028] Figure 2 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] The present application is further described below in conjunction with specific embodiments and drawings. It is to be understood that the illustrative embodiments of the present disclosure are only intended to explain the present application, rather than to limit the present application. In addition, for ease of description, only some, but not all, structures or processes related to the present application are shown in the drawings.

[0030] The following specific embodiments illustrate the implementation of the present application, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Although the description of the present application will be introduced in conjunction with the preferred embodiment, this does not mean that the features of this invention are limited to this implementation. On the contrary, the purpose of introducing the invention in conjunction with the implementation is to cover other options or modifications that may extend based on the claims of the present application. In order to provide an in-depth understanding of the present application, the following description will include many specific details. The present application can also be implemented without using these details. In addition, in order to avoid confusion or blurring the focus of the present application, some specific details will be omitted in the description.

[0031] Unless the context dictates otherwise, the terms "comprising," "having," and "including" are synonymous. The phrase "A / B" means "A or B." The phrase "A and / or B" means "(A and B) or (A or B)."

[0032] It should be understood that although the terms "first", "second", "one", "another", etc. may be used herein to describe various components, units, data, or things, these components, units, data, or things should not be limited by these terms. These terms are used only to distinguish one feature from another. For example, without departing from the scope of the exemplary embodiment, a first feature may be referred to as a second feature, and similarly a second feature may be referred to as a first feature. In addition, one may be referred to as another, and similarly another may be referred to as one.

[0033] It should be noted that in this specification, similar reference numerals and letters represent the same or similar items in the drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0034] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0035] The technical problem to be solved by this application is to provide a method for predicting the value of real estate, by constructing a real estate value assessment index system, constructing a Stacking fusion model for the index variables after feature selection, and performing precision weighted optimization, so as to achieve accurate and efficient prediction of the value of real estate. The method for predicting the value of real estate in this application includes the following specific steps.

[0036] Step S1, acquiring data information related to real estate to establish a first data set, and preprocessing the first data set, the first data set includes multiple independent variables and at least one dependent variable, and the data information related to real estate includes at least one of the following information: project information, location information, asset information, transaction description information, or matter description information.

[0037] Data information related to real estate includes multiple indicator dimensions, such as project information, location information, asset information, transaction description information, and event description information. They constitute the core elements that affect the value of real estate. The various elements are interrelated and work together to affect the final value prediction results. The specific information of the above indicator dimensions is shown in the following table.

[0038]

[0039] Step S1 also includes the following specific steps.

[0040] Step S11: Desensitize the first data set by using masking technology without changing the distribution characteristics of the first data set.

[0041] Without changing the original distribution characteristics of the data, sensitive information is replaced to protect data privacy and security. For sensitive information in real estate value prediction (such as personal identity information, specific address, etc.), masking technology is used for desensitization. The desensitization process can be expressed as:

[0042] X desensitized =f mask (X sensitive ) (1)

[0043] In formula (1), X sensitive is the original sensitive data, X desensitized is the anonymized data, f mask is the mask function.

[0044] Step S12: Use the K nearest neighbor algorithm to perform data filling on the first data set.

[0045] To solve the missing value problem in the data set, we use the K-nearest neighbor algorithm (KNN) to fill in the missing values ​​by finding the K complete data points that are most similar to the missing data points and using their medians. missing , calculate its difference with other complete data points x in the data set i (i=1,...,n), select the nearest K complete data points, and then calculate the median of these data points as the filling value. The formula is:

[0046]

[0047] In formula (2), is the distance x missing The value of the nearest i-th data point, K is the preset number of neighbors parameter.

[0048] Step S13: The multiple independent variables include categorical independent variables and numerical independent variables. Each categorical independent variable is converted into a set of binary vectors using one-hot encoding, and each numerical independent variable is scaled to the interval [0, 1] using range normalization.

[0049] In order to improve the model training efficiency and prediction accuracy, logarithmic transformation is used for the dependent variable to reduce the skewed distribution of the data, that is:

[0050] y′=log(y″) (3)

[0051] Where y″ is the original real estate value and y′ is the value after logarithmic transformation.

[0052] For independent variables, they are processed separately according to their data types. For categorical independent variables, one-hot encoding is used to convert each categorical independent variable into a set of binary vectors, each vector has only one position of 1 and the rest are 0, so as to ensure that the model can process non-numeric data. For numerical independent variables, the range standardization method is used to scale the numerical data to the interval [0, 1] to eliminate the dimension effect. The formula is:

[0053]

[0054] In the formula, x is the original value of the numerical independent variable, x′ is the standardized value, and x min and x max are the minimum and maximum values ​​of the numeric independent variable, respectively.

[0055] Separate the dependent variables and independent variables after data transformation, and divide the features in the data set D into the independent variable set X and the dependent variable y. Let X be an m×n matrix, where m is the number of samples and n is the number of independent variables; y is an m×1 vector, representing the corresponding dependent variable value, which can be expressed as:

[0056] X=[x′1,x′2,...,x′ n ] (5)

[0057] y=y′ (6)

[0058] In formula (5), the independent variable set X includes all features except the dependent variable, and in formula (6), the dependent variable y is the logarithm of the real estate value.

[0059] Step S2: Calculate the feature importance scores of the multiple independent variables based on the first random forest model, set a threshold according to the distribution of the feature importance scores, and delete the independent variables in the first data set whose feature importance scores are less than the threshold, so as to obtain the second data set.

[0060] The feature importance scores of the respective variables are calculated and ranked based on the first random forest model, a threshold S is set according to the distribution of the feature importance scores, and independent variables with scores exceeding the threshold S are retained. It should be noted here that those skilled in the art can set the threshold S according to actual conditions, such as the number of original independent variable data and the number of independent variables required for model training, and step S2 also includes the following specific steps.

[0061] Step S21, using a first random forest model to model multiple independent variables and at least one dependent variable, the first random forest model includes multiple decision trees, each decision tree randomly extracts at least one independent variable from the multiple independent variables for node splitting during training, so as to predict the importance of each independent variable;

[0062] The first random forest model is used to model the independent variable X and the dependent variable y. The first random forest model consists of multiple decision trees. During training, each decision tree randomly extracts a subset of features from X for node splitting to evaluate the importance of each feature to the prediction result. The model can be expressed as:

[0063] M RF ={T1, T2, ..., T k} (7)

[0064] In formula (7), T i is the i-th decision tree. The model training process can be expressed as:

[0065] M RF ←RandomForestRegressor(X,y,n estimators , random state ) (8)

[0066] In formula (8), n estimators is the number of decision trees, random state is a random seed to ensure the reproducibility of the results.

[0067] Step S22: For each decision tree, calculate the Gini impurity reduction or information gain of each independent variable at all split nodes in the decision tree, and then average the calculation results of all decision trees to obtain the feature importance score of each independent variable;

[0068] The feature importance score is evaluated based on the contribution of each feature in the decision tree split node. For each decision tree T i , calculate each feature x′ j The reduction in Gini impurity or information gain at all split nodes in a decision tree. The results of all decision trees are averaged to obtain for each feature x′ j The global importance score Imp(x′ j ), feature x′ j The importance score of can be expressed as:

[0069]

[0070] In formula (9), Importance(x′ j , T i ) is the feature x′ j In the i-th decision tree T i It should be pointed out here that in this application, independent variables and features represent two different ways of describing the same definition.

[0071] Step S23: sort the multiple independent variables in descending order according to the calculated feature importance scores, set a threshold according to the distribution of the feature importance scores, and delete the independent variables in the first data set whose feature importance scores are less than the threshold, so as to obtain the second data set.

[0072] According to the calculated feature importance score Imp(x′ j ), sort the independent variables in descending order to form a feature importance sequence:

[0073] F sorted =[x′1,x′2,...,x′ n ] sorted byImp (10)

[0074] According to the distribution of feature importance scores, the threshold is set to S, and the independent variables with scores greater than or equal to the threshold S are retained to form a screened independent variable set:

[0075] X selected ={x′ j |Imp(x′ j )≥S} (11)

[0076] In formula (11), S is the threshold, X selected The set of independent variables retained after feature selection, that is, the set of independent variables of the second data set, is used for subsequent model training and prediction.

[0077] Step S3, constructing five basic models, namely, a BP neural network model, a second random forest model, an XGBoost model, a LightGBM model and a CatBoost model, and using the TPOT (Tree-based Pipeline Optimization Tool) automated machine learning tool to optimize the hyperparameters of the five basic models respectively, selecting one of the basic models as a meta-learner in turn, and the remaining four basic models as base learners, using the second data set and using the Stacking algorithm to fuse the five basic models respectively, thereby constructing five Stacking fusion models, and selecting the optimal Stacking fusion model from the five Stacking fusion models through model performance evaluation, and the meta-learner corresponding to the optimal Stacking fusion model is the optimal meta-learner;

[0078] Based on the basic model after TPOT parameter tuning, the Stacking fusion model is constructed and the optimal meta-learner is determined.

[0079] Five basic models were constructed based on BP neural network model, second random forest model, XGBoost (eXtreme GradientBoosting) model, LightGBM (Light Gradient-Boosting Machine) model and CatBoost (Categorical Boosting) model, and hyperparameters were tuned using TPOT automatic machine learning technology. One of the basic models was selected as the meta-learner, and the other four basic models were selected as base learners. The five basic models were fused using the Stacking algorithm to find the optimal meta-learner to build the Stacking fusion model.

[0080] In this application, the above five basic models are selected to construct the Stacking fusion model, mainly based on comprehensive considerations such as model diversity and performance complementarity.

[0081] The above five basic models cover different types of machine learning algorithms, including neural network models (BP neural network model), tree-based ensemble methods (random forest model, XGBoost model, LightGBM model), and enhanced decision tree methods (CatBoost model) that focus on processing category features. Diversified models can capture different features and / or patterns in the data, thereby improving the generalization ability of the fusion model.

[0082] Moreover, each of the five basic models has its own unique advantages and applicable scenarios, and can complement each other in performance. For example, the BP neural network model is good at processing complex nonlinear relationships, and the random forest model can process high-dimensional data and is insensitive to outliers; the XGBoost model, LightGBM model, and CatBoost model achieve efficient training and prediction through gradient boosting technology, and the CatBoost model is particularly good at processing data with a large number of categorical features. The complementarity of these models enables them to complement each other when fused, improving the overall prediction performance.

[0083] It should also be noted that the first random forest model and the second random forest model are preferably two different random forest models constructed for different tasks. However, the present invention is not limited thereto, and the first random forest model and the second random forest model may also be the same model.

[0084] Step S3 also includes the following specific steps.

[0085] Divide the second data set into training set D train and the test set D test , suppose the second data set contains m samples, then the number of samples in the training set is m train , the number of test set samples is m test , and m train +m test =m.

[0086] Five basic models, including BP neural network model, random forest model, XGBoost model, LightGBM model and CatBoost model, are constructed as candidate base learners and meta-learners of the Stacking fusion model.

[0087] The BP neural network model is a multi-layer feedforward neural network that is trained using the back propagation algorithm. Its model can be expressed as:

[0088] y BPNN =f BPNN (X;θ BPNN ) (12)

[0089] In formula (12), f BPNN is the model mapping function, θ BPNN is the model parameter.

[0090] The random forest model is an ensemble learning model composed of multiple decision trees, and its output is the average of the outputs of each decision tree. Suppose the output of a single decision tree is y tree , its model can be expressed as:

[0091]

[0092] In formula (13), N is the number of decision trees.

[0093] The XGBoost model is a distributed gradient boosting library based on gradient boosting technology. Its model can be expressed as:

[0094]

[0095] In formula (14), f k is the function of the kth tree, θ k is the model parameter and K is the total number of trees.

[0096] The LightGBM model is a distributed gradient boosting framework based on decision trees, which uses a histogram-based algorithm to accelerate the training process. Its model output is similar to the XGBoost model, but it optimizes computational efficiency and memory usage. Its model can be expressed as:

[0097] y LightGBM =LightGBM(X;θ LightGBM ) (15)

[0098] In formula (15), LightGBM is the model mapping function, θ LightGBM is the model parameter.

[0099] The CatBoost model is a gradient boosting library that can handle categorical and numerical independent variables, and reduces prediction bias through a symmetric tree structure. The objective function also contains a loss function and a regularization term, but special treatment is given to categorical independent variables. The model can be expressed as:

[0100] y CatBost =CatBoost(X;θ CatBoost ) (16)

[0101] In formula (16), CatBoost is the model mapping function, θ CatBoost is the model parameter.

[0102] Using TPOT automatic machine learning technology, the five basic models are hyperparameter tuned. TPOT searches for the optimal model structure and hyperparameter combination through genetic algorithms. Its optimization process can be expressed as:

[0103] θ * = argmax θ∈Θ Performance(Model(X train ,y train ;θ)) (17)

[0104] In the formula, θ *is the optimal hyperparameter combination, Θ is the hyperparameter space, and Performance is the model performance evaluation indicator.

[0105] One of the five basic models is selected as the meta-learner in turn, and the remaining four basic models are used as base learners, and are fused through the Stacking algorithm. Specifically, for each basic model, its prediction result is input into the meta-learner as a new feature, and the meta-learner makes predictions based on these features.

[0106] For each base model (as a base learner), use the training set D train Train and test set D test The samples in are predicted and the prediction results are obtained The prediction results of the four base learners are used as new features and together with the independent variable X, form a new training set and test set Select one of the remaining base models as a meta-learner and use the new training set Train and test the new set Make a prediction and get the final prediction result y meta .

[0107] Repeat the above process, taking each basic model as a meta-learner and the remaining basic models as base learners in turn, and construct five Stacking fusion models, including BPNN-Stacking model, RF-Stacking model, XGBo0st-Stacking model, LightGBM-Stacking model, and CatBoost-Stacking model.

[0108] By performing performance evaluation, the Stacking fusion model with the best performance is selected, and its corresponding meta-learner is determined to be the optimal meta-learner.

[0109] The performance evaluation of the model can be carried out from several aspects such as prediction accuracy, operation efficiency and degree of deviation.

[0110] Prediction accuracy is a key indicator to measure the closeness between the model prediction value and the actual value. The coefficient of determination (R 2 ), mean square error (MSE), root mean square error (RMSE) or mean absolute error (MAE) as the evaluation criteria.

[0111] The coefficient of determination is used to measure the degree of fit between the model prediction value and the actual value. The closer its value is to 1, the better the model prediction effect is. The calculation formula is:

[0112]

[0113] In formula (18), is the model prediction value, is the actual value, is the mean of the actual values, m test is the number of test set samples.

[0114] The mean square error evaluates the average value of the sum of squares of the model prediction errors. The smaller the value, the smaller the model prediction error. The calculation formula is:

[0115]

[0116] The root mean square error is the square root of the mean square error, which has the same dimension as the original data, making it easier to intuitively understand the error size. The calculation formula is:

[0117]

[0118] The mean absolute error evaluates the average value of the absolute value of the model prediction error. The smaller the value, the smaller the model prediction error. The calculation formula is:

[0119]

[0120] Operational efficiency is an important indicator to measure the speed of model training and prediction, which is quantified by model training time and running time.

[0121] Training time records the total time required for the model to be trained from initialization to completion, including data preprocessing, model building, parameter tuning, and final training. Running time records the total time required for the model to make predictions on the test set, including data loading, model loading, and prediction result output.

[0122] By comparing the performance of the models on the above time indicators, we can evaluate the improvement of the model operation efficiency.

[0123] The degree of deviation is an important indicator to measure the difference between the model prediction results and the calculation results of traditional real estate valuation methods. By comparing the real estate value predicted by the model with the calculation results of traditional real estate valuation methods such as cost method, income method, and market method, the degree of deviation of the model prediction results is evaluated to verify the effectiveness and practicality of the model improvement.

[0124] Step S4: Optimize the optimal Stacking fusion model using the precision weighted strategy to obtain a real estate value prediction model.

[0125] On the basis of selecting the optimal meta-learner, considering that the traditional Stacking fusion model only averages the prediction results of the test set, it is easy to average out the training results of the model with good performance. Here, the precision weighted strategy is used to improve the traditional Stacking fusion model, that is, the prediction results of the four base learners are dynamically adjusted instead of simply taking the average.

[0126] Step S4 also includes the following specific steps.

[0127] Step S41: Use the second data set to perform five-fold cross validation. Specifically, the training set is randomly divided into five equal parts, four parts are selected as training subsets each time, and the remaining one part is used as validation subset, and five independent training and validation are performed in total. For each base learner i=1,2,3,4, based on the prediction results in the j-fold cross validation and the corresponding true value Calculate the mean square error:

[0128]

[0129] In formula (22), m valid is the number of samples in the validation subset,

[0130] Based on the mean squared error, calculate the accuracy of each base learner in the j-fold cross validation:

[0131] Accuracy i，j =1-MSE i，j (twenty three)

[0132] On this basis, for each base learner, calculate its average accuracy in five-fold cross validation:

[0133]

[0134] Step S42: In each training iteration, based on the average accuracy, determine the dynamic weight ω i :

[0135]

[0136] In each training iteration, the weights of each base learner are adjusted proportionally according to the change in accuracy on the validation set, so that the model with higher accuracy gets a larger weight. This mechanism ensures that the performance differences of the model on different subsets can be fully considered and effectively integrated into the subsequent prediction results.

[0137] Step S43: In the test phase, for each test sample, four base learners are used to perform predictions respectively to obtain prediction results. i=1,2,3,4, based on dynamic weight ωi , perform weighted average on the prediction results to obtain the weighted average prediction result:

[0138]

[0139] Step S44: Use the weighted average prediction result as new test data for testing the optimal meta-learner, optimize the optimal Stacking fusion model, and thus obtain a real estate value prediction model.

[0140] Figure 2 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application.

[0141] like Figure 2 As shown, in some embodiments, the electronic device 10 may include one or more processors 200, a system control logic 201 connected to at least one of the processors 200, a system memory 202 connected to the system control logic 201, a non-volatile memory (NVM) 203 connected to the system control logic 201, and a network interface 204 connected to the system control logic 201.

[0142] In some embodiments, the processor 200 may include one or more single-core or multi-core processors. In some embodiments, the processor 200 may include any combination of a general-purpose processor and a special-purpose processor (e.g., a graphics processor, an application processor, a baseband processor, etc.). For example, the processor 200 may include any combination of a general-purpose processor and a special-purpose processor required to implement the aforementioned real estate value prediction method.

[0143] In some embodiments, system control logic 201 may include any suitable interface controller to provide any suitable interface to at least one of processors 200 and / or any suitable device or component in communication with system control logic 201 .

[0144] In some embodiments, the system control logic 201 may include one or more memory controllers to provide an interface to the system memory 202. The system memory 202 may be used to load and store data and / or instructions. In some embodiments, the system memory 202 of the electronic device 10 may include any suitable volatile memory, such as a suitable dynamic random access memory (DRAM).

[0145] The non-volatile memory 203 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. In some embodiments, the non-volatile memory 203 may include any suitable non-volatile memory such as flash memory and / or any suitable non-volatile storage device, such as at least one of a HDD (hard disk drive), a CD (compact disc) drive, and a DVD (digital versatile disc) drive.

[0146] The non-volatile memory 203 may include a portion of a storage resource installed on a device of the electronic device 10 , or it may be accessible by the device but not necessarily a portion of the device. For example, the non-volatile memory 203 may be accessed over a network via the network interface 204 .

[0147] In particular, the system memory 202 and the non-volatile memory 203 / storage 516 may include: a temporary copy and a permanent copy of the instruction 205, respectively. The instruction 205 may include: when executed by at least one of the processors 200, the electronic device 10 may implement the following Figure 1 In some embodiments, instructions 205 , hardware, firmware and / or software components thereof may additionally / alternatively be placed in system control logic 201 , network interface 204 and / or processor 200 .

[0148] The network interface 204 may include a transceiver for providing a radio interface for the electronic device 10, thereby communicating with any other suitable device (such as a front-end module, an antenna, etc.) through one or more networks. In some embodiments, the network interface 204 may be integrated with other components of the electronic device 10. For example, the network interface 204 may be integrated with at least one of the processor 200, the system memory 202, the non-volatile memory 203, and a firmware device (not shown) having instructions. When at least one of the processors 200 executes the instructions, the electronic device 10 implements the above-mentioned Figure 1 The process shown.

[0149] The network interface 204 may further include any suitable hardware and / or firmware to provide a multiple-input multiple-output radio interface. For example, the network interface 204 may be a network adapter, a wireless network adapter, a telephone modem and / or a wireless modem.

[0150] In one embodiment, at least one of the processors 200 may be packaged with logic for one or more controllers of the system control logic 201 to form a system in package (SiP). In one embodiment, at least one of the processors 200 may be integrated on the same die with logic for one or more controllers of the system control logic 201 to form a system on chip (SoC).

[0151] The electronic device 10 may further include an input / output (I / O) device 206. The I / O device 206 may include a user interface to enable a user to interact with the electronic device 10; and a peripheral component interface design to enable peripheral components to interact with the electronic device 10.

[0152] In some embodiments, the user interface may include, but is not limited to, a display (eg, a liquid crystal display, a touch screen display, etc.), a sensor, a speaker, a microphone, a light emitting diode, and physical buttons.

[0153] In some embodiments, the peripheral component interface may include, but is not limited to, a non-volatile memory port, an audio jack, and a power interface.

[0154] In some embodiments, the sensor may include, but is not limited to, a gyroscope sensor, an acceleration sensor, a GNSS antenna, and a positioning unit. The positioning unit may also be a part of the network interface 204 or interact with the network interface 204 to communicate with components of the positioning network (e.g., GNSS satellites).

[0155] The embodiments of the present application also provide a computer program product for implementing the real estate value prediction method provided in the above embodiments.

[0156] The various embodiments of the mechanism disclosed in the present application can be implemented in hardware, software, firmware or a combination of these implementation methods. The embodiments of the present application can be implemented as a computer program module or module code executed on a programmable system, and the programmable system includes at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device and at least one output device.

[0157] A computer program module or module code can be applied to input instructions to perform the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, a processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0158] Module code can be implemented with high-level modular language or object-oriented programming language to communicate with the processing system. When necessary, module code can also be implemented with assembly language or machine language. In fact, the mechanism described in this application is not limited to the scope of any specific programming language. In either case, the language can be a compiled language or an interpreted language.

[0159] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, instructions may be distributed over a network or through other computer-readable media. Therefore, a machine-readable medium may include any mechanism for storing or transmitting information in a machine (e.g., computer) readable form, including, but not limited to, a floppy disk, an optical disk, an optical disk, a magneto-optical disk, a read-only memory (ROM), a random access memory (RAM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic card or an optical card, a flash memory, or a tangible machine-readable memory for transmitting information (e.g., a carrier wave, an infrared signal, a digital signal, etc.) using the Internet in an electrical, optical, acoustic, or other form of propagation signal. Accordingly, machine-readable media include any type of machine-readable media suitable for storing or transmitting electronic instructions or information in a form readable by a machine (eg, a computer).

[0160] References to "one embodiment" or "an embodiment" in the specification mean that the specific features, structures, or characteristics described in conjunction with the embodiment are included in at least one exemplary implementation or technology disclosed according to the embodiment of the present application. The appearance of the phrase "in one embodiment" in various places in the specification does not necessarily all refer to the same embodiment.

[0161] The disclosure of the embodiment of the present application also relates to an operating device for executing the text. The device can be specially constructed for the required purpose or it can include a general-purpose computer selectively activated or reconfigured by a computer program stored in the computer. Such a computer program can be stored in a computer-readable medium, such as, but not limited to any type of disk, including a floppy disk, an optical disk, a CD-ROM, a magneto-optical disk, a read-only memory (ROM), a random access memory (RAM), an EPROM, an EEPROM, a magnetic or optical card, an application-specific integrated circuit (ASIC) or any type of medium suitable for storing electronic instructions, and each can be coupled to a computer system bus. In addition, the computer mentioned in the specification may include a single processor or may be an architecture involving multiple processors for increased computing power.

[0162] The above-mentioned embodiments are only used to illustrate the present invention, in which the real estate value assessment index system, basic model, parameter tuning, etc. can all be changed. Any equivalent transformations or improvements based on the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A method for predicting the value of real estate, characterized in that: The steps include: Step S1, acquiring data information related to real estate to establish a first data set, and preprocessing the first data set, wherein the first data set includes multiple independent variables and at least one dependent variable, and the data information related to real estate includes at least one of the following information: project information, location information, asset information, transaction description information, or event description information; Step S2: calculating the feature importance scores of the plurality of independent variables based on the first random forest model, setting a threshold according to the distribution of the feature importance scores, and deleting the independent variables in the first data set whose feature importance scores are less than the threshold, thereby obtaining a second data set; Step S3, constructing five basic models, namely, a BP neural network model, a second random forest model, an XGBoost model, a LightGBM model and a CatBoost model, using the TPOT automated machine learning tool to optimize the hyperparameters of the five basic models respectively, selecting one of the basic models as a meta-learner in turn, and the remaining four basic models as base learners, using the second data set and using the Stacking algorithm to fuse the five basic models respectively, thereby constructing five Stacking fusion models, selecting the optimal Stacking fusion model from the five Stacking fusion models through model performance evaluation, and the meta-learner corresponding to the optimal Stacking fusion model is the optimal meta-learner; Step S4: Optimize the optimal Stacking fusion model using a precision weighted strategy to obtain a real estate value prediction model.

2. The method for predicting the value of real estate according to claim 1, characterized in that: The step S1 comprises: Step S11, desensitizing the first data set by using a masking technology without changing the distribution characteristics of the first data set; Step S12, using a K nearest neighbor algorithm to perform data filling on the first data set; Step S13, the multiple independent variables include categorical independent variables and numerical independent variables, each of the categorical independent variables is converted into a set of binary vectors using one-hot encoding, and each of the numerical independent variables is scaled to the interval [0, 1] using range normalization.

3. The method for predicting the value of real estate according to claim 2, characterized in that: The step S13 further includes: the at least one dependent variable is the real estate value, and a logarithmic transformation is performed on the dependent variable to reduce the skewed distribution.

4. The method for predicting the value of real estate according to claim 1, characterized in that: The step S2 comprises: Step S21, using a first random forest model to model the multiple independent variables and the at least one dependent variable, the first random forest model comprising a plurality of decision trees, each decision tree randomly extracting at least one independent variable from the multiple independent variables for node splitting during training, thereby predicting the importance of each independent variable; Step S22: For each decision tree, calculate the Gini impurity reduction or information gain of each independent variable at all split nodes in the decision tree, and then average the calculation results of all decision trees to obtain the feature importance score of each independent variable; Step S23: sort the multiple independent variables in descending order according to the calculated feature importance scores, set a threshold according to the distribution of the feature importance scores, and delete the independent variables in the first data set whose feature importance scores are less than the threshold, so as to obtain a second data set.

5. The method for predicting the value of real estate according to claim 1, characterized in that: The step S4 further comprises: Step S41: Perform five-fold cross validation using the second data set. For each base learner i=1, 2, 3, 4, based on the prediction result in the j-fold cross validation and the corresponding true value Calculate the mean square error: In the formula, m valid is the number of samples in the validation subset, Based on the mean square error, calculate the accuracy of each base learner in the j-fold cross validation: Accuracy i,j =1-MSE i,j For each base learner, calculate its average accuracy in the five-fold cross validation: Step S42: In each training iteration, based on the average accuracy, determine the dynamic weight ω i : Step S43: In the test phase, for each test sample, four base learners are used to perform predictions respectively to obtain prediction results. Based on the dynamic weight ω i , perform weighted averaging on the prediction results to obtain a weighted average prediction result: Step S44: using the weighted average prediction result as new test data for testing the optimal meta-learner, optimizing the optimal Stacking fusion model, and thereby obtaining a real estate value prediction model.

6. An electronic device, characterized in that: include: one or more processors; One or more memories; the one or more memories store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device executes the real estate value prediction method described in any one of claims 1 to 5.

7. A computer-readable medium, characterized in that The readable medium stores instructions, which, when executed on a computer, cause the computer to execute the method for predicting the value of real estate according to any one of claims 1 to 5.

8. A computer program product, characterized in that It comprises a computer program / instruction, which, when executed by a processor, implements the method for predicting the value of real estate according to any one of claims 1 to 5.

Images