WO2018151343A1

WO2018151343A1 - System and method for automatically assessing real estate price on basis of ensemble learning technique

Info

Publication number: WO2018151343A1
Application number: PCT/KR2017/001642
Authority: WO
Inventors: 맹준영; 이창노; 최우현; 장명수; 김경주; 김정은
Original assignee: 맹준영; 이창노; 최우현; 장명수; 김경주; 김정은
Priority date: 2017-02-15
Filing date: 2017-02-15
Publication date: 2018-08-23

Abstract

The present invention comprises: a real estate database in which real estate-specific actual transaction price, geographical coordinates, and similar price zone segment data are stored in advance; a price estimation model design module, which applies a parametric model and a nonparametric model to the actual price, the geographical coordinates, and similar price zone segment data of the corresponding real estate so as to respectively calculate a model-specific price for predetermined real estate of a population, assigns weights, which are the same or different from each other, to each calculated model-specific price so as to calculate a weight-applied price, compares the calculated weight-applied price to the price zone segment data of corresponding real estate so as to determine similarity therebetween such that the price difference is determined according to physical location, and interpolating the weight-applied price in a space in consideration of physical location so as to calculate an estimated price, when the determination result price difference according to the physical location of the weight-applied price is greater than or equal to a predetermined reference value; an estimated price validation verification module for verifying the validation of the estimated price by comparing the estimated price with the actual transaction price; a population price estimation module for calculating the estimated price of the remaining real estates in the population according to the calculation method of the price estimation model design module when the validity is verified; an estimated price adequacy review module for reviewing the adequacy of the estimated price of the remaining real estate in the population by comparing the price range of the estimated price of the population real estate and the price range of the actual transaction price; and an estimated price provision module for providing the reviewed estimated price online when the estimated price is reviewed as being appropriate.

Description

Automated Real Estate Price Evaluation System and Method Based on Ensemble Learning Technique

The present invention relates to an automatic real estate price evaluation system and method, and more particularly, to a real estate price automatic evaluation system and method based on an ensemble learning technique.

In the past, parametric models such as linear regression models were mainly used to estimate or determine real estate prices. The parameter model has a lot of strict assumptions such as the normality of data, the independence of explanatory variables, and the linearity of the price function, which limits the reliability of the estimated price.

There are various nonlinear factors that cannot be determined by the parameter model alone, but this cannot be reflected by existing models. For example, as the land area increases, the transaction price tends to decrease, but the parameter model cannot reflect these characteristics due to the linearity assumption.

In addition, conventionally, after applying various models to data, only one model having the best prediction result was selected and adopted as the final model. However, several estimation models have their own characteristics, which can not be reflected in the same context as the aforementioned problem.

In addition, real estate prices vary greatly depending on their physical location, which is called spatial dependence. Conventionally, the characteristics of the real estate price formation, that is, the factors such as physical location and distance have not been reflected, and the accuracy in estimating the real estate price is inevitably deteriorated.

An object of the present invention to provide an automatic real estate price evaluation system based on the ensemble learning technique.

Another object of the present invention is to provide an automatic real estate price evaluation method based on an ensemble learning technique.

The automatic real estate price evaluation system based on the ensemble learning method according to the above object of the present invention includes a real estate database in which real transaction prices, geographic coordinates, and similar price zone division data are stored in advance; At least one parametric model and at least one non-parametric model are applied to the real estate, geographic coordinates, and pseudo-price parity data of the real estate previously stored in the real estate database. Calculate the weighted price by calculating the price for each model for each model, and assigning the same or different weights to the calculated price for each model, wherein the calculated weighted price is compared to the similar price range partition data of the property. The price difference according to the physical location is determined by the similarity, and if the price difference according to the physical location of the weighted price is more than a predetermined reference value, the weighted price is estimated by interpolating the space in consideration of the physical location. Price estimation model to calculate the price System module; An estimated price validation module for validating the estimated price against the estimated price compared with the actual transaction price stored in the real estate database; A population price estimation module that calculates an estimated price of the remaining real estate of the population according to a calculation method of the price estimation model design module, when the validation is validated in the estimated price validation module; Estimated price adequacy review module that examines the adequacy of the estimated price of the remaining real estate of the population against the price range of the estimated price of the real estate estimated by the population price estimation module and the price range of the actual transaction price stored in the real estate database ; When the estimated price is reviewed as appropriate in the estimated price adequacy review module, the estimated price adequacy module may be configured to include an estimated price providing module that provides the estimated estimated price online.

The price estimating model design module may be configured to apply a linear regression model among the parameter models and to apply a random forest model or a boosting model among the non-parametric models.

In addition, the price estimating model design module applies an arithmetic mean weight to the price of each model or uses an artificial neural network model to calculate an optimal weight for each model according to the reliability of the price of each model. It can be configured to calculate and apply.

In addition, the price estimating model design module linearly combines the actual transaction price of the real estate in consideration of the corresponding physical location of the real estate in which the transaction has already been stored in the real estate database, and spaces the weighted price using the linear combined actual transaction price. Can be configured to calculate an estimated price by means of a kriging technique that interpolates over.

Meanwhile, the estimated price validation module calculates a price ratio, a variance coefficient, and an absolute average error rate by comparing the estimated price with an actual transaction price stored in the real estate database, and verifies the estimated price according to the calculation result. Can be configured.

The real estate price automatic evaluation method based on the ensemble learning method according to another object of the present invention, the price estimation model design module is at least one or more to the actual price, geographic coordinates and similar price zone partition data pre-stored in the real estate database Calculating a model-specific price for a given real estate of a population by applying a parametric model and at least one non-parametric model; Calculating, by the price estimating model design module, a weighted price by assigning the same or different weights to each of the calculated prices for each model; Determining, by the price estimating model design module, a price difference according to a physical location based on whether the calculated weighted price is similar to the similar price zone segment data of the real estate; Calculating the estimated price by interpolating the weighted price in space in consideration of the physical location when the price difference according to the physical location of the weighted price is greater than or equal to a predetermined reference value as a result of the determination; A step of validating the estimated price by the estimated price validation module comparing the estimated price with an actual transaction price stored in the real estate database; When the estimated price validation module is validated, calculating, by the population price estimation module, an estimated price of the remaining real estate of the population according to a calculation method of the price estimation model design module; The estimated price adequacy review module examines the adequacy of the estimated prices of the remaining properties of the population against the price range of the estimated price of the real estate estimated by the population price estimation module and the price range of the actual transaction price stored in the real estate database. Doing; When the estimated price is reviewed as being appropriate in the estimated price adequacy review module, the estimated price providing module may be configured to include providing the reviewed estimated price online.

In this case, the price estimating model design module applies at least one parameter model and at least one non-parametric model to the actual price, geographic coordinates, and pseudo-price parity data of the real estate previously stored in the real estate database, and then models the predetermined real estate of the population. The calculating of each price may be configured to apply a linear regression model among the parameter models and to apply a random forest model or a boosting model among the non-parametric models.

The calculating of the weighted price by assigning the same or different weights to the calculated price for each model by the price estimation model design module may apply an arithmetic mean weight to the price of each model or for each model. According to the reliability of the price, an artificial neural network model can be configured to calculate and apply an optimal weight for each model price.

In this case, when the price difference according to the physical location of the weighted price is more than a predetermined reference value, the price estimation model design module calculates an estimated price by interpolating the weighted price in space in consideration of the physical location. Krigging for linearly combining the actual transaction price of the real estate in consideration of the corresponding physical location of the real estate that has already occurred transaction stored in the real estate database and interpolating the weighted price in space using the linear combined actual transaction price The technique may be configured to calculate an estimated price.

Meanwhile, the validating of the estimated price by comparing the estimated price with the actual transaction price stored in the real estate database may include comparing the estimated price with the actual transaction price stored in the real estate database. The price ratio, the variance coefficient, and the absolute mean error rate may be calculated, and the validity of the estimated price may be verified according to the calculation result.

According to the system and method for automatically evaluating real estate prices based on the ensemble learning technique described above, by applying optimal weights to various models by an artificial neural network algorithm, the estimations of various models are optimized and reflected, and the accuracy of estimation is improved. In addition, by estimating the real estate price in consideration of the physical location of the real estate, it is effective to reflect an important factor of the formation of the real estate price and increase the accuracy of estimation.

1 is a block diagram of an automatic real estate price evaluation system based on an ensemble learning method according to an embodiment of the present invention.

Figure 2 is an illustration of an artificial neural network model according to an embodiment of the present invention.

3 is an exemplary diagram illustrating a result of spatial interpolation according to an embodiment of the present invention.

4 is an exemplary view showing a validation result of a price estimation model according to an embodiment of the present invention.

5 is an exemplary diagram of an online providing screen of an estimated price according to an embodiment of the present invention.

6 is a flowchart of a method for automatically evaluating real estate prices based on an ensemble learning method according to an embodiment of the present invention.

As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing the drawings, similar reference numerals are used for similar elements.

Terms such as first, second, A, and B may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. The term and / or includes a combination of a plurality of related items or any item of a plurality of related items.

When a component is referred to as being "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in between. Should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that there is no other component in between.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Referring to FIG. 1, an automatic real estate price evaluation system based on an ensemble learning method according to an embodiment of the present invention (hereinafter, referred to as an “real estate price automatic evaluation system”) 100 is a real estate database 110 and a price estimation model. It may be configured to include the design module 120, the estimated price validation module 130, the population price estimation module 140, the estimated price adequacy review module 150, and the estimated price providing module 160.

Hereinafter, the detailed structure is demonstrated.

The real estate database 110 may be configured to store property-specific characteristics, actual transaction prices, geographic coordinates, and similar price range partition data in advance. Here, the similar price zone partition data may be configured to be databased in advance as partition data forming a similar price zone according to a physical location. For example, if the price range is high in the backcountry zone, it means data about the area around the station where the advantage of the backyard zone applies.

The price estimation model design module 120 is a component for designing an optimal price estimation model by applying various models for estimating real estate prices.

The price estimating model design module 120 first generates a new price estimating model by applying weights to various price estimating models, and performs the interpolation according to the physical location to reflect the change characteristics of real estate prices according to the physical location. Can be. Specifically, it is as follows.

The price estimating model design module 120 includes at least one parametric model and at least one non-parametric model in the real price, geographic coordinates, and pseudo-price parcel data of the property previously stored in the real estate database 110. A parametric model) may be applied to calculate a price for each model for a given real estate.

Here, the linear regression model is applied to the parameter model, and the random forest model, boosting model, support vector machine, or multivariate adaptive regression splines are applied to the non-parametric model. Can be. The non-parametric model is highly realistic because it estimates real estate prices without strict assumptions.

The price estimating model design module 120 combines and applies these price estimating models. The price estimating model design module 120 applies arithmetic mean weights that give the same weights to the prices of each model, or according to the reliability of the price of each model. An artificial neural network model may be used to calculate and apply an optimal weight for each model price to calculate a weighted price. In the case of applying the artificial neural network model, an ensemble learning technique in which different weights are applied to the price of each model is applied. Here, the ensemble learning technique can cancel out different tendencies such as overestimation or underestimation.

Meanwhile, the price estimation model design module 120 cancels the price difference according to the physical location with respect to the weighted price. First, the weighted price is configured to determine whether the weighted price is similar to the similar price zone segment data of the property. Can be.

The price estimation model design module 120 may be configured to determine a price difference according to the physical location of the property through the determination result. As a result, when the price difference according to the physical location of the weighted price is more than a predetermined reference value, the price estimation model design module 120 may be configured to calculate the estimated price by interpolating the weighted price in space in consideration of the physical location. have.

Here, the price estimating model design module 120 may be configured to calculate the estimated price by interpolating the weighted price by a kriging technique.

Specifically, the price estimating model design module 120 linearly combines the actual transaction prices of the real estate in consideration of the corresponding physical location of the real estate in which the transaction has already been stored in the real estate database 110 and weights using the linear combined actual transaction prices. The application price may be configured to interpolate in space.

The estimated price validation module 130 may be configured to validate the estimated price. In this case, the estimated price validation module 130 may be configured to verify the validity of the estimated price by comparing the estimated price with the actual transaction price stored in the real estate database 110. Diagnostic indicators for validating may include a price ratio for determining the accuracy of the estimated price, a coefficient of dispersion for determining the balance, and a mean absolute percentage error. That is, the estimated price validation module 130 calculates a price ratio, a variance coefficient, and an absolute average error rate by comparing the estimated price with the actual transaction price stored in the real estate database 110, and verifies the validity of the estimated price according to the calculation result. It can be configured to.

The population price estimation module 140 may be configured to calculate the estimated price of the remaining real estate of the population according to the calculation method of the price estimation model design module 120 when the validation is performed in the estimated price validation module 130. . At this time, the remaining real estate is real estate that is not traded among the population.

The estimated price adequacy review module 150 estimates the remaining real estate of the population by comparing the price range of the estimated price of the real estate estimated by the population price estimation module 140 with the price range of the actual transaction price stored in the real estate database 110. It may be configured to review the adequacy of the price.

Here, the indicator for examining the adequacy of the estimated price through the combined model may include the average unit price, the minimum unit price and the maximum unit price of the actual transaction price included in the real estate database 110. That is, the estimated price adequacy review module 150 may be configured to review the adequacy of the average unit price, the minimum unit price, and the maximum unit price of the actual transaction price with respect to the estimated price.

If the estimated price adequacy review module 150 considers the estimated price to be appropriate, the estimated price providing module 160 may be configured to provide the estimated price reviewed online online.

Referring to FIG. 2, the artificial neural network model is a method of linearly combining explanatory variables in various ways, and then predicting the dependent variable using the linear combination in the form of a nonlinear function. Hereinafter, a single hidden layer and a feedforward neural network model will be described. However, this description is equally extended to more complex neural network models such as multiple hidden layers.

Dependent Variables in Neural Network Model

Is a nonlinear function

, Where

Is m derived predictors

It is composed of the following forms.

Equation 1

Normal

Will be assigned to

(jth derived explanatory variable of the i observation) is a nonlinear function

Expressed as

Is the first explanatory variable

It consists of a linear combination of. The formula is as follows.

Equation 2

Combining Equations 1 and 2 above is equal to Equation 3 below.

Equation 3

M functions in the above expression

Is called an activation function, and its typical form is the following logistic function. This type of function is known to have the flexibility to fit relatively well in various situations.

Equation 4

Equation 3 is re-expressed using an activation function as in Equation 5 below.

Equation 5

here,

Is a parameter,

Is an explanatory variable,

Is the residual.

The logistic activation function has a value between 0 and 1, so if the dependent variable is continuous

You need to convert it so that takes a value between 0 and 1.

Equation 6

Neural network models usually represent their structure in the form of a network. A typical linear regression model and a neural network model are shown in FIG. 2.

In FIG. 2, the middle H represents m hidden nodes in the right neural network model. Activation function

Identity function, not a logistic function

We assume that the neural network model is simplified to a general linear regression model. In other words, the neural network model can be interpreted as a generalization of the linear regression model. The process of finding parameter values in such a neural network model is often performed through a procedure called 'back-propagation', and the present invention may be configured to estimate parameters according to this method.

In FIG. 3, when houses are traded as shown in the left figure, the values of all the houses not traded are spatially interpolated based on the transaction prices of these houses as shown in the right figure. In other words, spatial interpolation is the construction of the right surface data (areal data) based on the point data on the left.

4 is a calculation of the estimated price ratio by the model compared to the actual transaction price of the Amsa-dong housing, the overall effectiveness of the model can be determined by looking at the high and low price ratio.

If it is determined that the estimate is appropriate, it may be configured to provide the estimated price such as FIG. 5 online.

Referring to FIG. 6, the price estimation model design module 120 may include at least one parametric model and at least one of the real price, geographic coordinates, and pseudo-price parity data of the real estate previously stored in the real estate database 110. By applying the above non-parametric model (S101) to calculate the price for each model for a given real estate of the population (S101).

In this case, the price estimation model design module 120 may be configured to apply a linear regression model among the parameter models, and apply a random forest model or a boosting model among the non-parametric models.

Next, the price estimation model design module 120 calculates the weighted price by giving the same or different weights to the calculated price for each model, respectively (S102).

Here, the price estimation model design module 120 applies an arithmetic mean weight to the price of each model or optimizes the price of each model using an artificial neural network model according to the reliability of the price of each model. It may be configured to calculate and apply a weight.

Next, the price estimation model design module 120 determines the price difference according to the physical location based on whether the weighted price calculated above is similar to the similar price zone partition data of the property (S103).

If the price difference according to the physical location of the weighted price is greater than or equal to a predetermined reference value, the price estimation model design module 120 calculates an estimated price by interpolating the weighted price in space in consideration of the physical location (S104). .

At this time, the price estimation model design module 120 linearly combines the actual transaction price of the real estate in consideration of the physical location of the real estate that has already occurred transaction stored in the real estate database 110 and apply a weight using the linear combined actual transaction price It can be configured to calculate an estimated price by a kriging technique that interpolates the price in space.

Next, the estimated price validation module 130 verifies the validity of the estimated price by comparing the estimated price with the actual transaction price stored in the real estate database 110 (S105).

In this case, the estimated price validation module 130 calculates a price ratio, a variance coefficient, and an absolute average error rate by comparing the estimated price with the actual transaction price stored in the real estate database 110, and verifies the validity of the estimated price according to the calculation result. It can be configured to.

Next, when the validity is verified in the estimated price validation module 130, the population price estimation module calculates an estimated price of the remaining real estate of the population according to the calculation method of the price estimation model design module 120 (S106).

Next, the estimated price adequacy review module 150 compares the price range for the estimated price of the real estate estimated by the population price estimation module 140 with the price range of the actual transaction price stored in the real estate database 110. Review the appropriateness of the estimated price of the real estate (S107).

Next, when the estimated price is reviewed as appropriate in the estimated price adequacy review module 150, the estimated price providing module 160 provides the reviewed estimated price online (S108).

Although described with reference to the above embodiments, those skilled in the art can understand that the present invention can be variously modified and changed without departing from the spirit and scope of the invention described in the claims below. There will be.

Claims

A real estate database in which real transaction prices, geographic coordinates, and pseudo-price parcel data by real estate are stored in advance;

At least one parametric model and at least one non-parametric model are applied to the real estate, geographic coordinates, and pseudo-price parity data of the real estate previously stored in the real estate database. Calculate the weighted price by calculating the price for each model for each model, and assigning the same or different weights to the calculated price for each model, wherein the calculated weighted price is compared to the similar price range partition data of the property. The price difference according to the physical location is determined by the similarity, and if the price difference according to the physical location of the weighted price is more than a predetermined reference value, the weighted price is estimated by interpolating the space in consideration of the physical location. Price estimation model to calculate the price System module;

An estimated price validation module for validating the estimated price against the estimated price compared with the actual transaction price stored in the real estate database;

A population price estimation module that calculates an estimated price of the remaining real estate of the population according to a calculation method of the price estimation model design module, when the validation is validated in the estimated price validation module;

Estimated price adequacy review module that examines the adequacy of the estimated price of the remaining real estate of the population against the price range of the estimated price of the real estate estimated by the population price estimation module and the price range of the actual transaction price stored in the real estate database ;

And if the estimated price is reviewed as appropriate in the estimated price adequacy review module, an automatic price estimation system based on an ensemble learning technique including an estimated price providing module that provides the estimated estimated price online.
The method of claim 1, wherein the price estimation model design module,

An automatic real estate price evaluation system based on an ensemble learning method, wherein the linear regression model is applied among the parameter models, and a random forest model or a boosting model is applied among the non-parametric models.
The method of claim 1, wherein the price estimation model design module,

It is configured to apply the arithmetic mean weight to the price of each model or to calculate and apply the optimal weight for each model using artificial neural network model according to the reliability of the price of each model. Real estate price automatic evaluation system based on the ensemble learning technique.
The method of claim 1, wherein the price estimation model design module,

In the kriging technique of linearly combining the actual transaction price of the real estate in consideration of the corresponding physical location of the real estate in which the transaction has already occurred stored in the real estate database and interpolating the weighted price in space using the linear combined actual transaction price. Real estate price automatic valuation system based on an ensemble learning technique, characterized in that it is configured to calculate an estimated price.
The method of claim 1, wherein the estimated price validation module,

An ensemble learning technique configured to calculate a price ratio, a variance coefficient, and an absolute mean error rate by comparing the estimated price with an actual transaction price stored in the real estate database, and verify the validity of the estimated price according to the calculation result. Based real estate price automatic evaluation system.
The price estimating model design module applies at least one parametric model and at least one non-parametric model to the actual price, geographic coordinates, and pseudo-price parcel data of the property previously stored in the real estate database. Calculating a price for each model of a predetermined real estate of the population;

Calculating, by the price estimating model design module, a weighted price by assigning the same or different weights to each of the calculated prices for each model;

Determining, by the price estimating model design module, a price difference according to a physical location based on whether the calculated weighted price is similar to the similar price zone segment data of the real estate;

Calculating the estimated price by interpolating the weighted price in space in consideration of the physical location when the price difference according to the physical location of the weighted price is greater than or equal to a predetermined reference value as a result of the determination;

A step of validating the estimated price by the estimated price validation module comparing the estimated price with an actual transaction price stored in the real estate database;

When the estimated price validation module is validated, calculating, by the population price estimation module, an estimated price of the remaining real estate of the population according to a calculation method of the price estimation model design module;

The estimated price adequacy review module examines the adequacy of the estimated prices of the remaining properties of the population against the price range of the estimated price of the real estate estimated by the population price estimation module and the price range of the actual transaction price stored in the real estate database. Doing;

And when the estimated price is reviewed as being appropriate in the estimated price adequacy review module, providing an estimated price online by the estimated price providing module, based on an ensemble learning technique.
7. The method of claim 6, wherein the price estimating model design module applies at least one parameter model and at least one non-parametric model to the actual price, geographic coordinates, and pseudo-price parity data of the property previously stored in the real estate database. Computing the price of each model for real estate,

Applying a linear regression model of the parameter model, and random forest model or boosting model of the non-parametric model is configured to apply the real estate price automatic evaluation method based on the ensemble learning method.
The method of claim 6, wherein the price estimating model design module calculates a weighted price by assigning weights equal to or different from each other to the calculated price for each model,

It is configured to apply the arithmetic mean weight to the price of each model or to calculate and apply the optimal weight for each model using artificial neural network model according to the reliability of the price of each model. Real estate price automatic evaluation method based on the ensemble learning technique.
According to claim 6, If the result of the determination that the price difference according to the physical location of the weighted price is more than a predetermined reference value, the price estimation model design module interpolates the weighted price in space in consideration of the physical location to estimate the estimated price Calculating the step,

In the kriging technique of linearly combining the actual transaction price of the real estate in consideration of the corresponding physical location of the real estate in which the transaction has already occurred stored in the real estate database and interpolating the weighted price in space using the linear combined actual transaction price. Automatic price estimation method based on an ensemble learning technique, characterized in that it is configured to calculate the estimated price.
The method of claim 6, wherein the estimating price validation module validates the estimated price by comparing the estimated price with an actual transaction price stored in the real estate database.

An ensemble learning technique configured to calculate a price ratio, a variance coefficient, and an absolute mean error rate by comparing the estimated price with an actual transaction price stored in the real estate database, and verify the validity of the estimated price according to the calculation result. Automated Valuation of Real Estate Prices Based.