Automatically determining a current value for a home

• US 10,074,111 B2
• Filed: 01/29/2014
• Issued: 09/11/2018
• Est. Priority Date: 02/03/2006
• Status: Active Grant

• Current Assignee: MFTB Holdco Incorporated
• Original Assignee: Zillow, Inc. (Zillow Group, Inc.)
• Inventors: Humphries, Stanley B., Xiang, Dong, Chung, Kyusik, Burstein, Jonathan Lee
• Primary Examiner(s): Tran, Hai
• Application Number:
  US14/167,962
• Publication Number:
  US 20140236845A1
• Time in Patent Office: 1,686 Days
• Field of Search: 705 35- 45
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0278 Product appraisal
• CPC Class Codes: G06Q 50/16 Real estate

Abstract

A facility for valuing a distinguished home located in a distinguished geographic area is described. The facility receives home attributes for the distinguished home. The facility obtains valuation for the distinguished home by applying to the received home attributes evaluation model for homes in the distinguished geographic area that has been trained using selling price and home attribute data from homes recently sold in the distinguished geographic area. The facility reports the obtained valuation for the distinguished home.

Claims

1. A method, in a computing system having a memory and a processor, for valuing a distinguished home located in a distinguished geographic area, comprising:

retrieving, by the processor, home sales data for the distinguished geographic area from a plurality of source systems over a computer network, the home sales data comprising a plurality of entries each indicating, for a home located in the distinguished geographic area that was recently sold, a selling price, and, for each of a plurality of attributes, the value of the attribute for the home;

creating, by the processor, a plurality of classification trees for the distinguished geographic area;

storing each of the created plurality of classification trees for the distinguished area in the memory;

for each of the classification trees, by the processor;

randomly selecting a proper subset of the plurality of entries;

randomly selecting a proper subset of the plurality of attributes;

for each of the selected attributes, determining the full range of values of the selected attribute among the selected entries;

establishing a root node representing all of the selected entries and the full range of values of each of the selected attributes;

for each node of the tree;

determining the information gain borne by each possible split of each of the ranges of the selected attributes represented by the node to the selling prices of the entries represented by the node;

when the greatest information gain of a possible split exceeds the information gain of the node;

performing the possible split having the greatest information gain to divide the range into two subranges at a point in the attribute range that produces the largest variance between an average selling price for the subranges to an average selling price for the range;

for each of the two subranges, establishing a child of the node representing the subrange and the homes represented by the node whose attribute values fall into the subrange;

when the greatest information gain of a possible split does not exceed the information gain of the node, identifying the node as a leaf node and calculating a mean selling price for the homes represented by the node;

for each of a proper subset of the plurality of entries that excludes the selected entries;

identifying a leaf node of the classification tree representing attribute ranges containing the entry'"'"'s attributes;

comparing the price of the identified leaf node to the selling price of the entry;

scoring the classification tree based on the extent to which the prices of the identified leaf nodes differed from the corresponding selling prices;

receiving, by the processor, attributes of the distinguished home from a user device over the computer network;

for each of the classification trees, identifying, by the processor, a leaf node of the classification tree representing attribute ranges containing the distinguished home'"'"'s attributes;

determining, by the processor, an average of the price of the identified leaf node in each of the trees that is weighted by the tree'"'"'s score; and

reporting the determined average as the value of the distinguished home to the user device.

2. A computer-readable medium not constituting transitory signals whose contents cause a computing system to perform a method for valuing a distinguished home located in a distinguished geographic area in cooperation with a memory, the method comprising:

receiving home attributes for the distinguished home from a client device over a computer network;

storing the received home attributes in the memory;

obtaining, with the processor, a valuation for the distinguished home by applying to the home attributes stored in the memory a weighted classification tree-based valuation model, stored in the memory, for homes in the distinguished geographic area trained using selling price and home attribute data from homes recently sold in the distinguished geographic area,

wherein the valuation model is a compound model that includes a component for all homes in the distinguished geographic area, as well as a component for the most highly-valued homes in the distinguished geographic area; and

reporting, with the processor, the obtained valuation for the distinguished home to the client device.

3. The computer-readable medium not constituting transitory signals of claim 2 wherein the received home attributes are retrieved from a database of public information.

4. The computer-readable medium not constituting transitory signals of claim 2 wherein the received home attributes are inputted by a person knowledgeable about the attributes of the distinguished home.

5. The computer-readable medium not constituting transitory signals of claim 2 wherein the compound valuation model is applied by first applying the component for all homes in the distinguished geographic area, and using the produced valuation to weight valuations generated for the home by the two components in the reported valuation.

6. The computer-readable medium not constituting transitory signals of claim 2, further comprising:

determining that a home attribute value for the distinguished home is unavailable; and

in response to the determination, imputing a value for the attribute for the distinguished home.

7. The computer-readable medium not constituting transitory signals of claim 6, further comprising choosing as the imputed attribute value the median value of the attribute among homes recently sold in the distinguished geographic area.

8. The computer-readable medium not constituting transitory signals of claim 6, further comprising choosing as the imputed attribute value the mode value of the attribute among homes recently sold in the distinguished geographic area.

9. The computer-readable medium not constituting transitory signals of claim 2, the method further comprising, before reporting the obtained valuation for the distinguished home, blending into the obtained valuation an earlier-reported valuation for the distinguished home.

10. The computer-readable medium not constituting transitory signals of claim 9 wherein the blending comprises generating a weighted average of the obtained valuation and the earlier-reported valuation in which the earlier-reported valuation is more heavily weighted than the obtained valuation.

11. The computer-readable medium not constituting transitory signals of claim 2 wherein the valuation model is comprised of a first constituent valuation model trained using selling price and home attribute data from homes recently sold in the distinguished geographic area at the highest selling prices and a second constituent valuation model trained using selling price and home attribute data from homes recently sold in the distinguished geographic area at all selling prices, and wherein obtaining a valuation for the distinguished home comprises blending the constituent valuations obtained by applying each of the constituent valuation models to the received home attributes.

12. The computer-readable medium not constituting transitory signals of claim 2 wherein the valuation model is comprised of a first constituent valuation model trained using selling price and home attribute data from homes recently sold in the distinguished geographic area at the highest selling prices and a second constituent valuation model trained using selling price and home attribute data from homes recently sold in the distinguished geographic area at all selling prices, and wherein obtaining a valuation for the distinguished home comprises:

applying the second constituent model to obtain the second constituent valuation;

if the obtained second constituent valuation is below a threshold percentile among selling prices in the geographic area, using the second constituent valuation as the valuation;

if the second constituent valuation is above the threshold percentile selling price in the geographic area;

applying the first constituent valuation model to obtain a first constituent valuation; and

computing a weighted average of the constituent valuations in which the weighting is based upon the extent to which the second constituent valuation exceeds the threshold percentile selling price in the geographic area to obtain the valuation.

13. A method, in a computing system having a processor and a memory, for establishing a valuation model for homes in a distinguished geographic area, comprising:

receiving, by the processor, sales information for homes in the distinguished geographic area sold after a distinguished past date from a plurality of source systems over a computer network, the sales information including home attributes and selling price for each home;

storing the received sales information in the memory;

initializing, by the computer system, a valuation model, stored in the memory, that is a forest of classification trees; and

training, by the computer system, the initialized valuation model with the sales information stored in the memory for at least a portion of the homes in the distinguished geographic area sold after the distinguished past date, such that the trained valuation model values each of these homes at or near its selling price, wherein each of classification trees stored in the memory is weighted based upon a level of success of the classification tree in valuing homes in the distinguished geographic area that were recently sold other than those used to train the valuation model.

14. The method of claim 13 wherein the valuation model is trained using the identified sales information for a first proper subset of homes in the distinguished geographic area sold after the distinguished past date, the method further comprising:

selecting a second proper subset of homes in the distinguished geographic area sold after the distinguished past date that is distinct from the first proper subset; and

using identified sales information for the second proper subset of homes to validate the trained valuation model by determining the extent to which the trained valuation model'"'"'s valuation of the homes of the second proper subset diverges from the selling prices of the homes of the second proper subset.

15. The method of claim 13, further comprising:

identifying among the homes in the distinguished geographic area sold after the distinguished past date homes that have undergone unusually rapid appreciation or depreciation; and

excluding the identified homes from the training of the valuation model.

16. The method of claim 15 wherein homes are identified whose change in selling price from second most recent sale to most recent sale lies outside a predetermined percentage range.

17. The method of claim 16, further comprising choosing as the imputed attribute value the median value of the attribute among homes in the distinguished geographic area sold after the distinguished date.

18. The method of claim 16, further comprising choosing as the imputed attribute value the mode value of the attribute among homes in the distinguished geographic area sold after the distinguished date.

19. The method of claim 15, further comprising:

selecting from the homes in the distinguished geographic area sold after the distinguished past date homes whose change in selling price from second most recent sale to most recent sale lies within a first predetermined percentage range;

initializing and training a second valuation model with identified sales information for the selected homes; and

applying the second valuation model to all of the homes in the distinguished geographic area sold after the distinguished past date, and wherein homes are identified whose difference between selling price from second most recent sale to valuation obtained by applying the second valuation model lies outside a second predetermined percentage range.

20. The method of claim 13, further comprising:

identifying a home in the distinguished geographic area sold after the distinguished past date having an attribute for which no value is available; and

in response to the identifying, imputing a value of the attribute for the identified home.

21. The method of claim 13, further comprising:

identifying a home in the distinguished geographic area sold after the distinguished past date for which no selling price is available; and

in response to the identifying, imputing a selling price for the identified home.

22. The method of claim 21, further comprising:

choosing as the imputed selling price the product of a loan amount and an estimated regional home-to-value ratio.

23. One or more computer memories, not constituting transitory signals, collectively storing:

a forest of classification tree data structures usable to value homes in a distinguished geographic area, each of the trees of the forest classifying a random subset of homes and their recent selling prices based on values of a random subset of known home attributes,

such that the forest may be used to value a home in the distinguished geographic area based upon its home attributes,

wherein associated with each tree of the forest is a weight indicating the likely level of accuracy of values produced by the tree relative to the other trees.

24. A method, in a computing system having a processor and a memory, of valuing a distinguished home located in a distinguished geographic area, comprising:

receiving, by the processor, home attributes for the distinguished home from a plurality of source systems over a computer network;

storing the home attributes in the memory;

obtaining, by a processor, a valuation for the distinguished home by applying to the home attributes stored in the memory a weighted classification tree-based valuation model for homes in the distinguished geographic area trained using selling price and home attribute data from homes recently sold in the distinguished geographic area, wherein the valuation model is a compound model that includes a component for all homes in the distinguished geographic area, as well as a component for the most highly-valued homes in the distinguished geographic area; and

reporting the obtained valuation for the distinguished home to a client device over the computer network.

25. A computing system for establishing a valuation model for homes in a distinguished geographic area, comprising:

a processor and memory, together comprising;

an identifying unit configured to identify sales information for homes in the distinguished geographic area sold after a distinguished past date from a plurality of source systems across a computer network, the sales information including home attributes and selling price for each home;

an initializing unit configured to initialize a valuation model that is a forest of classification trees; and

a training unit configured to train the initialized valuation model with the identified sales information for at least a portion of the homes in the distinguished geographic area sold after the distinguished past date, such that the trained valuation model values each of these homes at or near its selling price,

wherein each of classification trees is weighted based upon a level of success of the classification tree in valuing homes in the distinguished geographic area that were recently sold other than those used to train the valuation model.