Automatically determining a current value for a home

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.

Description

TECHNICAL FIELD

The described technology is directed to the field of electronic commerce techniques, and, more particularly, to the field of electronic commerce techniques relating to real estate.

BACKGROUND

In many roles, it can be useful to be able to accurately determine the value of residential real estate properties (“homes”). As examples, by using accurate values for homes: taxing bodies can equitably set property tax levels; sellers and their agents can optimally set listing prices; buyers and their agents can determine appropriate offer amounts; insurance firms can properly value their insured assets; and mortgage companies can properly determine the value of the assets securing their loans.

A variety of conventional approaches exist for valuing houses. Perhaps the most reliable is, for a house that was very recently sold, attributing its selling price as its value. Unfortunately, following the sale of a house, its current value can quickly diverge from its sale price. Accordingly, the sale price approach to valuing a house tends to be accurate for only a short period after the sale occurs. For that reason, at any given time, only a small percentage of houses can be accurately valued using the sale price approach.

Another widely-used conventional approach to valuing houses is appraisal, where a professional appraiser determines a value for a house by comparing some of its attributes to the attributes of similar nearby homes that have recently sold (“comps”). The appraiser arrives at an appraised value by subjectively adjusting the sale prices of the comps to reflect differences between the attributes of the comps and the attributes of the house being appraised. The accuracy of the appraisal approach can be adversely affected by the subjectivity involved. Also, appraisals can be expensive, can take days or weeks to completed, and may require physical access to the house by the appraiser.

In view of the shortcomings of conventional approaches to valuing houses discussed above, a new approach to valuing houses that was more universally accurate, less expensive, and more convenient would have significant utility.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram showing some of the components typically incorporated in at least some of the computer systems and other devices on which the facility executes.

FIG. 2 is a flow diagram showing steps typically performed by the facility to automatically determine current values for homes in a geographic area.

FIG. 3 is a table diagram showing sample contents of a recent sales table.

FIG. 4A is a flow diagram showing steps typically performed by the facility in order to construct a tree.

FIG. 4B is a flow diagram showing steps typically performed by the facility in order to determine whether and how to split a node of a tree.

FIG. 5 is a table diagram showing sample contents of a basis table containing the basis information selected for the tree.

FIG. 6 is a tree diagram showing a root node corresponding to the basis table 500.

FIG. 7 is a tree diagram showing a completed version of the sample tree.

FIG. 8 shows steps typically performed by the facility in order to score a tree.

FIG. 9 is a table diagram showing sample results for scoring a tree.

FIG. 10 is a display diagram showing detailed information about an individual home.

FIG. 11 is a display diagram showing a map identifying a number of homes in the same geographic area.

DETAILED DESCRIPTION

A software facility for automatically determining a current value for a home (“the facility”) is described. In some embodiments, the facility establishes, for each of a number of geographic regions, a model of housing prices in that region. This model transforms inputs corresponding to home attributes into an output constituting a predicted current value of a home in the corresponding geographic area having those attributes. In order to determine the current value of a particular home, the facility selects the model for a geographic region containing the home, and subjects the home's attributes to the selected model.

In some embodiments, the facility constructs and/or applies housing price models each constituting a forest of classification trees. In some such embodiments, the facility uses a data table that identifies, for each of a number of homes recently sold in the geographic region to which the forest corresponds, attributes of the home and its selling price. For each of the trees comprising the forest, the facility randomly selects a fraction of homes identified in the table, as well as a fraction of the attributes identified in the table. The facility uses the selected attributes of the selected homes, together with the selling prices of the selected homes, to construct a classification tree in which each non-leaf node represents a basis for differentiating selected homes based upon one of the selected attributes. For example, where number of bedrooms is a selected attribute, a non-leaf node may represent the test “number of bedrooms ≤4.” This node defines 2 subtrees in the tree: one representing the selected homes having 4 or fewer bedrooms, the other representing the selected homes having 5 or more bedrooms. Each leaf node of the tree represents all of the selected homes having attributes matching the ranges of attribute values corresponding to the path from the tree's root node to the leaf node. The facility assigns each leaf node a value corresponding to the mean of the selling prices of the selected homes represented by the leaf node.

In some areas of the country, home selling prices are not public records, and may be difficult or impossible to obtain. Accordingly, in some embodiments, the facility estimates the selling price of a home in such an area based upon loan values associated with its sale and an estimated loan-to-value ratio.

In order to weight the trees of the forest, the facility further scores the usefulness of each tree by applying the tree to homes in the table other than the homes that were selected to construct the tree, and, for each such home, comparing the value indicated for the home by the classification tree (i.e., the value of the leaf node into which the tree classifies the home) to its selling price. The closer the values indicated by the tree to the selling prices, the higher the score for the tree.

In most cases, it is possible to determine the attributes of a home to be valued. For example, they can often be obtained from existing tax or sales records maintained by local governments. Alternatively, a home's attributes may be inputted by a person familiar with them, such as the owner, a listing agent, or a person that derives the information from the owner or listing agent. In order to determine a value for a home whose attributes are known, the facility applies all of the trees of the forest to the home, so that each tree indicates a value for the home. The facility then calculates an average of these values, each weighted by the score for its tree, to obtain a value for the home. In various embodiments, the facility presents this value to the owner of the home, a prospective buyer of the home, a real estate agent, or another person interested in the value of the home or the value of a group of homes including the home.

In some embodiments, the facility applies its model to the attributes of a large percentage of homes in a geographic area to obtain and convey an average home value for the homes in that area. In some embodiments, the facility periodically determines an average home value for the homes in a geographic area, and uses them as a basis for determining and conveying a home value index for the geographic area.

Because the approach employed by the facility to determine the value of a home does not rely on the home having recently been sold, it can be used to accurately value virtually any home whose attributes are known or can be determined. Further, because this approach does not require the services of a professional appraiser, it can typically determine a home's value quickly and inexpensively, in a manner generally free from subjective bias.

FIG. 1 is a block diagram showing some of the components typically incorporated in at least some of the computer systems and other devices on which the facility executes. These computer systems and devices 100 may include one or more central processing units (“CPUs”) 101 for executing computer programs; a computer memory 102 for storing programs and data—including data structures, database tables, other data tables, etc.—while they are being used; a persistent storage device 103, such as a hard drive, for persistently storing programs and data; a computer-readable media drive 104, such as a CD-ROM drive, for reading programs and data stored on a computer-readable medium; and a network connection 105 for connecting the computer system to other computer systems, such as via the Internet, to exchange programs and/or data—including data structures. In various embodiments, the facility can be accessed by any suitable user interface including Web services calls to suitable APIs. While computer systems configured as described above are typically used to support the operation of the facility, one of ordinary skill in the art will appreciate that the facility may be implemented using devices of various types and configurations, and having various components.

FIG. 2 is a flow diagram showing steps typically performed by the facility to automatically determine current values for homes in a geographic area. The facility may perform these steps for one or more geographic areas of one or more different granularities, including neighborhood, city, county, state, country, etc. These steps may be performed periodically for each geographic area, such as daily. In step 201, the facility selects recent sales occurring in the geographic area. The facility may use sales data obtained from a variety of public or private sources.

FIG. 3 is a table diagram showing sample contents of a recent sales table. The recent sales table 300 is made up of rows 301-315, each representing a home sale that occurred in a recent period of time, such as the preceding 60 days. Each row is divided into the following columns: an identifier column 321 containing an identifier for the sale; an address column 322 containing the address of the sold home; a square foot column 323 containing the floor area of the home; a bedrooms column 324 containing the number of bedrooms in the home; a bathrooms column 325 containing the number of bathrooms in the home; a floors column 326 containing the number of floors in the home; a view column 327 indicating whether the home has a view; a year column 328 showing the year in which the house was constructed; a selling price column 329 containing the selling price at which the home was sold; and a date column 330 showing the date on which the home was sold. For example, row 301 indicates that sale number 1 of the home at 111 Main St., Hendricks, Ill. 62012 having a floor area of 1850 square feet, 4 bedrooms, 2 bathrooms, 2 floors, no view, built in 1953, was for $132,500, and occurred on Jan. 3, 2005. While the contents of recent sales table 300 were included to pose a comprehensible example, those skilled in the art will appreciate that the facility can use a recent sales table having columns corresponding to different and/or a larger number of attributes, as well as a larger number of rows. Attributes that may be used include, for example, construction materials, cooling technology, structure type, fireplace type, parking structure, driveway, heating technology, swimming pool type, roofing material, occupancy type, home design type, view type, view quality, lot size and dimensions, number of rooms, number of stories, school district, longitude and latitude, neighborhood or subdivision, tax assessment, attic and other storage, etc. For a variety of reasons, certain values may be omitted from the recent sales table. In some embodiments, the facility imputes missing values using the median value in the same column for continuous variables, or the mode (i.e., most frequent) value for categorical values.

While FIG. 3 and each of the table diagrams discussed below show a table whose contents and organization are designed to make them more comprehensible by a human reader, those skilled in the art will appreciate that actual data structures used by the facility to store this information may differ from the table shown, in that they, for example, may be organized in a different manner; may contain more or less information than shown; may be compressed and/or encrypted; etc.

Returning to FIG. 2, in steps 202-205, the facility constructs and scores a number of trees, such as 100. This number is configurable, with larger numbers typically yielding better results but requiring the application of greater computing resources. In step 203, the facility constructs a tree. In some embodiments, the facility constructs and applies random forest valuation models using an R mathematical software package available at http://cran.r-project.org/ and described at http://www.maths.lth.se/help/R/.R/library/randomForest/html/random Forest.html. Step 203 is discussed in greater detail below in connection with FIG. 4. In step 204, the facility scores the tree constructed in step 203. Step 204 is discussed in greater detail below in connection with FIG. 8.

In steps 206-207, the facility uses the forest of trees constructed and scored in steps 202-205 to process requests for home valuations. Such requests may be individually issued by users, or issued by a program, such as a program that automatically requests valuations for all homes in the geographic area at a standard frequency, such as daily, or a program that requests valuations for all of the homes occurring on a particular map in response to a request from a user to retrieve the map. In step 206, the facility receives a request for valuation identifying the home to be valued. In step 207, the facility applies the trees constructed in step 203, weighted by the scores generated for them in step 204, to the attributes in the home identified in the received request in order to obtain a valuation for the home identified in the request. After step 207, the facility continues in step 206 to receive the next request.

Those skilled in the art will appreciate that the steps shown in FIG. 2 and in each of the flow diagrams discussed below may be altered in a variety of ways. For example, the order of the steps may be rearranged; substeps may be performed in parallel; shown steps may be omitted, or other steps may be included; etc.

FIG. 4A is a flow diagram showing steps typically performed by the facility in order to construct a tree. In step 401, the facility randomly selects a fraction of the recent sales in the geographic area to which the tree corresponds, as well as a fraction of the available attributes, as a basis for the tree.

FIG. 5 is a table diagram showing sample contents of a basis table containing the basis information selected for the tree. Basis table 500 contains rows randomly selected from the recent sales table 300, here rows 302, 308, 209, 311, 313, and 315. The basis table further includes the identifier column 321, address column 322, and selling price column 329 from the recent sales table, as well as randomly selected columns for two available attributes: a bedrooms column 324 and a view column 327. In various embodiments, the facility selects various fractions of the rows and attribute columns of the recent sales table for inclusion in the basis table; here, the fraction one third is used for both.

In some embodiments, the facility filters rows from the basis table having selling prices that reflect particularly rapid appreciation or depreciation of the home relative to its immediately-preceding selling price. For example, in some embodiments, the facility filters from the basis table recent sales whose selling prices represent more than 50% annual appreciation or more than 50% annual depreciation. In other embodiments, however, the facility initially performs the filtering described above, then uses the filtered basis table to construct a preliminary model, applies the preliminary model to the unfiltered basis table, and excludes from the basis table used to construct the primary model those sales where the valuation produced by the preliminary model is either more than 2 times the actual selling price or less than one-half of the actual selling price.

Returning to FIG. 4A, in step 402, the facility creates a root node for the tree that represents all of the basis sales contained in the basis table and the full range of each of the basis attributes.

FIG. 6 is a tree diagram showing a root node corresponding to the basis table 500. The root node 601 represents the sales having identifiers 2, 8, 9, 11, 13, and 15; values of the bedrooms attribute between 1-∞; and values of the view attribute of yes and no.

Returning to FIG. 4A, in steps 403-407, the facility loops through each node of the tree, including both the root node created in step 402 and any additional nodes added to the tree in step 405. In step 404, if it is possible to “split” the node, i.e., create two children of the node each representing a different subrange of an attribute value range represented by the node, then the facility continues in step 405, &se the facility continues in step 406. FIG. 4B is a flow diagram showing steps typically performed by the facility in order to determine whether and how to split a node of a tree. These steps generally identify a potential split opportunity having the highest information gain, and determine whether the information gain of that potential split opportunity exceeds the information gain of the current node. In step 451, the facility determines whether the node's population—that is, the number of basis sales represented by the node—satisfies a split threshold, such as a split threshold that requires more than three basis sales. If the threshold is not satisfied, then the facility returns to step 404 in step 452 without identifying any split opportunity, such that the facility will not split the node; otherwise, the facility continues in step 453. Though not shown, the facility may apply a variety of other tests to determine whether the node should be split, including whether any of the selected attribute ranges represented by the node is divisible. For example, where the selected attributes are bedrooms and view, and a node represents the ranges bedrooms=5 and view=no, none of the node's selected attribute ranges can be split.

In steps 453-455, the facility analyzes the characteristics of the node in order to be able to compare them to characteristics of pairs of possible child nodes that would result from different opportunities for splitting the node. In step 453, the facility determines the mean selling price among the sales represented by the node to obtain a node mean selling price for the node. Applying step 453 to root node 600 shown in FIG. 6, the facility determines a mean selling price for the node as shown below in Table 1 by determining the mean of all the selling prices shown in basis table 500.

TABLE 1
1	Node mean selling price =	$201,400

In step 454, the facility sums the squares of the differences between the node mean selling price determined in step 454 and the selling price of each sale represented by the node to obtain a node overall squared error. This calculation is shown below in table 2 for root node 601.

TABLE 2
2	Sate 2 overall squared error = ($201,000-line 1 )2 =	160000
3	Sale 8 overall squared error = ($74,900-line 1)2 =	16002250000
4	Sale 9 overall squared error = ($253,500-line 1)2 =	2714410000
5	Sate 11 overall squared error = ($230,000-line 1)2 =	817960000
6	Sate 13 overall squared error = ($211,000-line 1)2 =	92160000
7	Sate 15 overall squared error = ($238,000-line 1)2 =	1339560000
8	Node overall squared error =	20966500000

In step 455, the facility divides the overall squared error by one fewer than the number of sales represented by the node in order to obtain a node variance. The calculation of step 455 for root node 600 is shown below in table 3.

TABLE 3
9	Node variance = line 8/5 =	4193300000

In steps 456460, the facility analyzes the characteristics of each possible split opportunity that exists in the node; that is, for each attribute range represented by the node, any point at which that range could be divided. For root node 600, three such split opportunities exist: (1) view=no/view=yes; (2) bedrooms≤4/bedrooms>4; and (3) bedrooms≤5/bedrooms>5. In step 457, for each side of the possible split opportunity, the facility determines the mean selling price among sales on that side to obtain a split side mean selling price. Table 4 below shows the performance of this calculation for both sides of each of the three possible split opportunities of root node 600.

TABLE 4
10	Split side mean selling price of view = no side of	$179,225
	possible split opportunity 1 = mean of selling prices for
	sales 2, 8, 11, and 13 =
11	Split side mean selling price of view = yes side of	$245,750
	possible split opportunity 1 = mean of selling prices for
	sales 9 and 15 =
12	Split side mean selling price for bedrooms ≤4 side of	$152,450
	possible split opportunity 2 = mean of selling prices of
	sales 8 and 11 =
13	Split side mean selling price for bedrooms >4 side of	$225,875
	possible split opportunity 2 = mean of selling prices of
	sales 2, 9, 13, and 15 =
14	Split side mean selling price for bedrooms ≤5 side of	$188,475
	possible split opportunity 3 = mean of selling prices of
	sales 8, 11, 13, and 15 =
15	Split side mean selling price for bedrooms >5 side of	$227,250
	possible split opportunity 3 = mean of selling prices of
	sales 2 and 9 =

In step 458, the facility sums the squares of the differences between the selling price of each sale represented by the node and the split side mean selling price on the same side of the possible split opportunity to obtain a possible split opportunity squared error. The result of the calculation of step 458 for root node 600 is shown below in table 5.

TABLE 5
16	Possible split opportunity 1 squared error for sale 2 =	474150625
	($201,000 − line 10)2 =
17	Possible split opportunity 1 squared error for sale 8 =	10883705625
	($74,900 − line 10)2 =
18	Possible split opportunity 1 squared error for sale 9 =	60062500
	($253,500 − line 11 )2 =
19	Possible split opportunity 1 squared error for sale 11 =	2578100625
	($230,000 − line 10)2 =
20	Possible split opportunity 1 squared error for sale 13 =	1009650625
	($211,000 − line 10)2 =
21	Possible split opportunity 1 squared error for sale 15 =	60062500
	($238,000 − line 11 )2 =
22	Possible split opportunity 1 squared error = sum of lines	15065732500
	16-21 =
23	Possible split opportunity 2 squared error for sale 2 =	618765625
	($201,000 − line 13)2 =
24	Possible split opportunity 2 squared error for sale 8 =	6014002500
	($74,900 − line 12)2 =
25	Possible split opportunity 2 squared error for sale 9 =	763140625
	($253,500 − line 13)2 =
26	Possible split opportunity 2 squared error for sale 11 =	6014002500
	($230,000 − line 12)2 =
27	Possible split opportunity 2 squared error for sale 13 =	221265625
	($211,000 − line 13)2 =
28	Possible split opportunity 2 squared error for sale 15 =	147015625
	($238,000 − line 13)2 =
29	Possible split opportunity 2 squared error = sum of lines	13778192500
	23-28 =
30	Possible split opportunity 3 squared error for sale 2 =	689062500
	($201,000 − line 15)2 =
31	Possible split opportunity 3 squared error for sale 8 =	12899280625
	($74,900 − line 14)2 =
32	Possible split opportunity 3 squared error for sale 9 =	689062500
	($253,500 − line 15)2 =
33	Possible split opportunity 3 squared error for sale 11 =	1724325625
	($230,000 − line 14)2 =
34	Possible split opportunity 3 squared error for sale 13 =	507375625
	($211,000 − line 14)2 =
35	Possible split opportunity 3 squared error for sale 15 =	2452725625
	($238,000 − line 14)2 =
36	Possible split opportunity 3 squared error = sum of lines	18961832500
	30-35 =

In line 459, the facility divides the possible split opportunity squared error by two less than the number of sales represented by the node to obtain a variance for the possible split opportunity. The calculation of step 459 is shown below for the three possible split opportunities of root node 600.

TABLE 6
37	Variance for possible split opportunity 1 = line 22/4 =	3766433125
38	Variance for possible split opportunity 2 = line 29/4 =	3444548125
39	Variance for possible split opportunity 3 = line 36/4 =	4740458125

In step 460, if another possible split opportunity remains to be processed, then the facility continues in step 456 to process the next possible split opportunity, else the facility continues in step 461.

In step 461, the facility selects the possible split opportunity having the lowest variance. In the example, the facility compares lines 37, 38 and 39 to identify the possible split opportunity 2 as having the lowest variance. In step 462, if the selected possible split opportunity variance determined in step 461 is less than the node variance determined in step 455, then the facility continues in step 464 to return, identifying the split opportunity selected in step 461, else the facility continues in step 463 to return without identifying a split opportunity. In the example, the facility compares line 38 to line 9, and accordingly determines to split the root node in accordance with split opportunity 2.

Returning to FIG. 4A, in step 405, where the steps shown in FIG. 4B determine that the node should be split, the facility creates a pair of children for the node. Each child represents one of the subranges of the split opportunity identified in step 404 and the node's full range of unselected attributes. Each child represents all basis sales whose attributes satisfy the attribute ranges represented by the child. Step 405 is discussed in greater detail below in connection with FIG. 7.

In step 406, because the node will be a leaf node, the facility determines the mean selling price of basis sales represented by the node.

In step 407, the facility processes the next node of the tree. After step 407, these steps conclude.

FIG. 7 is a tree diagram showing a completed version of the sample tree. It can be seen that the facility added child nodes 702 and 703 to root node 601, corresponding to the subranges defined by the split opportunity selected in step 461. Node 702 represents sales whose bedrooms attribute is less than or equal to 4, that is, between 1 and 4, as well as the full range of view attribute values represented by node 601. Accordingly, node 702 represents sales 8 and 11. Because this number of sales is below the threshold of 4, node 702 qualifies as a leaf node, and its valuation of $152,450 is calculated by determining the mean selling price of sales 8 and 11.

Node 703 represents sales with bedrooms attribute values greater than 4, that is, 5-∞. Node 703 further represents the full range of view attributes values for node 601. Accordingly, node 703 represents sales 2, 9, 13, and 15. Because this number of sales is not smaller than the threshold number and the node's ranges are not indivisible, the facility proceeded to consider possible split opportunities. In order to do so, the facility performs the calculation shown below in Table 7. For the following two possible split opportunities: (4) view=no/view=yes, and (5) bedrooms=5/bedrooms>5.

TABLE 7
40	node mean selling price - mean of selling prices for	$225,875
	sales 2, 9, 13, and 15 =
41	sale 2 overall squared error =	618765625
	($201,000 - line 40)2 =
42	sate 9 overall squared error =	76314625
	($253,500 - line 40)2 =
43	sale 13 overall squared error =	221265625
	($211,000 - line 40)2 =
44	sale 15 overall squared error =	147015625
	($238,000 - line 40)2 =
45	node overall squared error =	1750187500
46	node variance = line 45/3 =	583395833
47	split side mean selling price of view = no side of possible	$206,000
	split opportunity 4 = mean selling prices of sales 2 and 13 =
48	split side mean selling price of view-yes side of possible	$245,750
	split opportunity 4 = mean selling prices of sates 9 and 15 =
49	split side mean selling price for bedrooms ≤5 side of possible	$224,500
	split opportunity 5 = mean selling prices of sates 13 and 15 =
50	split side mean selling price of bedrooms>5 side of possible	$227,250
	split opportunity 5 = mean selling prices of sales 2 and 9 =
51	possible split opportunity 4 squared error for sale 2 =	25000000
	($201,000 - line 47)2 =
52	possible split opportunity 4 squared error for sale 9 =	60062500
	($253,500 - line 48)2 =
53	possible split opportunity 4 squared error for sale 13 =	25000000
	($211,000 - line 47)2 =
54	possible split opportunity 4 squared error for sate 15 =	60062500
	($238,000 - line 48)2 =
55	possible split opportunity 4 squared error = sum of lines	17012500
	51-54 =
56	possible split opportunity 5 squared error for sale 2 =	689062500
	($201,000 - line 50)2 =
57	possible split opportunity 5 squared error for sale 9 =	689062500
	($253,500 - line 50)2 =
58	possible split opportunity 5 squared error for sale 13 =	182250000
	($211,000 - line 49)2 =
59	possible split opportunity 5 squared error for sale 15 =	182250000
	($238,000 - line 49)2 =
60	possible split opportunity 5 squared error = sum of lines	1742625000
	56-59 =
61	variance for possible split opportunity 4 = line 55/2 =	85062500
62	variance for possible split opportunity 5 = line 60/2 =	871312500

From Table 7, it can be seen that, between split opportunities 4 and 5, split opportunity 4 has the smaller variance, shown on line 61. It can further be seen that the variance of possible split opportunity 4 shown on line 61 is smaller than the node variance shown on line 46. Accordingly, the facility uses possible split opportunity 4 to split node 703, creating child nodes 704 and 705. Child node 704 represents basis sales 2 and 13, and that attribute ranges bedrooms=5-∞ and view=no. Node 704 has a valuation of $206,000, obtained by averaging the selling prices of the base of sales 2 and 13. Node 705 represents base of sales 9 and 15, and attribute value ranges bedrooms=5-∞ and view=yes. Node 705 has valuation $245,750, obtained by averaging the selling price of sales 9 and 15.

In order to apply the completed tree 700 shown in FIG. 7 to obtain its valuation for a particular home, the facility retrieves that home's attributes. As an example, consider a home having attribute values bedrooms=5 and view=yes. The facility begins at root node 601, and among edges 711 and 712, traverses the one whose condition is satisfied by the attributes of the home. In the example, because the value of the bedroom's attribute for the home is 5, the facility traverses edge 712 to node 703. In order to proceed from node 703, the facility determines, among edges 713 and 714, which edge's condition is satisfied. Because the home's value of the view attribute is yes, the facility traverses edge 714 to leaf node 705, and obtains a valuation for the sample home of $245,750.

Those skilled in the art will appreciate that the tree shown in FIG. 7 may not be representative in all respects of trees constructed by the facility. For example, such trees may have a larger number of nodes, and/or a larger depth. Also, though not shown in this tree, a single attribute may be split multiple times, i.e., in multiple levels of the tree.

FIG. 8 shows steps typically performed by the facility in order to score a tree. In step 801, the facility identifies recent sales in the geographic area that were not used as a basis for constructing the tree in order to score the tree. In steps 802-805, the facility loops through each sale identified in step 801. In step 803, the facility applies the tree to the attributes of the sale to obtain a value. In step 804, the facility compares the value obtained in step 803 to the selling price for the sale to determine an error magnitude, dividing the difference between valuation and selling price by selling price. In step 806, the facility calculates a score that is inversely related to the median error magnitude determined in step 804. After step 806, these steps conclude.

FIG. 9 is a table diagram showing sample results for scoring a tree. Scoring table 900 scores tree 700 based upon the contents of recent sales table 300. The scoring table is made up of the rows of recent sales table 300 other than those used as basis sales for constructing the tree, i.e., rows 301, 303, 304, 305, 306, 307, 310, 312, and 314. It further contains the following columns from recent sales table 300: identifier column 321, address column 322, bedroom column 324, view column 327, and selling price column 329. The scoring table further contains a valuation column 911 containing the valuation of each home determined in step 803. For example, row 307 shows that the facility determines the valuation of $245,750 for sale 7 using tree 700. In particular, the facility begins at root node 601; traverses to node 703 because the number of bedrooms 5 is greater than 4; traverses to node 705 because view=yes; and adopts the valuation of node 705, $245,750, Scoring table 900 further contains an error column 912 indicating the difference between each home's valuation and selling price. For example, row 307 contains an error of 0.0685, the difference between valuation $245,750 and selling price $230,000, divided by selling price $230,000. Associated with the table is a median error field 951 containing the median of error values in the scoring table, or 0.3734. Each tree's median error value is used to determine weightings for the trees that are inversely related to their median error values. In some embodiments, the facility determines the particular tree's weighting by generating an accuracy metric for each tree by subtracting its median error value from 1, and dividing the tree's accuracy measure by the sum of all of the trees' accuracy measures. Also, a variety of different approaches to determine a score that is negatively correlated with the average error may be used by the facility.

When a home is valued using the forest, the sample tree will be applied to the attributes of the home in the same way it was applied to homes in the scoring process described above. (If any attributes of the home are missing, the facility typically imputes a value for the missing attribute based upon the median or mode for that attribute in the recent sales table.) The valuation produced will be averaged with the valuations produced by the other trees of the forest. In the average, each valuation will be weighted by the score attributed by the facility to the tree. This resultant average is presented as the valuation for the home.

FIGS. 10-11 show ways in which valuations generated by the facility may be presented. FIG. 10 is a display diagram showing detailed information about an individual home. The display 1000 includes detailed information 1001 about the home. Despite the fact that the home has not been sold recently, the facility also displays a valuation 1002 for the home, enabling prospective buyers and listing agents to gauge their interest in the home, or permitting the home's owner to gauge his interest in listing the home for sale.

FIG. 11 is a display diagram showing a map identifying a number of homes in the same geographic area. The display 1100 shows homes 1101-1112. The facility also displays its valuations 1151-1162 of these homes in connection with their location on the map. Presenting the facility's valuations in this way permits home shoppers to obtain an overview of the geographic area, identify special trends within the geographic area, identify the anomalous values as good values or poor picks, etc.

In some embodiments, the valuations displayed or otherwise reported by the facility are not the “raw” valuations directly produced by the valuation model, but rather “smoothed” valuations that are generated by blending the raw valuation generated by the current iteration of the model with earlier valuations. As one example, in some embodiments, the facility generates a current smoothed valuation for a home by calculating a weighted average of a current raw valuation and a smoothed valuation of the same home from the immediately-preceding time period, where the prior smooth valuation is weighted more heavily than the current raw valuation. In some embodiments, where new iterations of the model are constructed and applied daily, the prior smoothed valuation is weighted 49 times as heavily as the current raw valuation; where a new iteration of the model is constructed and applied weekly, the prior smoothed valuation is weighted 9 times as heavily as the current raw valuation; where new iterations of the model are constructed and applied monthly, the previous smoothed valuation is weighted twice as heavily as the current raw valuation. Those skilled in the art will appreciate that a variety of other smoothing techniques may be used in order to dampen erratic movement in a particular home's reported valuation over time.

In some embodiments, the facility constructs and applies compound valuation models to one or more geographic areas. A compound valuation model includes two or more separate classification tree forests, some or all of which may be applied to the attributes of a particular home in order to value it. As one example, in some embodiments, the facility constructs a compound model including both a forest constructed as described above (referred to as a “core forest”), as well as a separate, “high-end” forest constructed from basis sales having a selling price above the 97.5 percentile selling price in the geographic area. In these embodiments, the compound model is applied as follows. First, the core forest is applied to the attributes of a home. If the valuation produced by the core forest is no larger than the 97.5 percentile selling price in the geographic area, then this valuation is used directly as the model's valuation. Otherwise, the facility also applies the high-end forest to the attributes of the home. If the valuation produced by the core forest is above the 99 percentile selling price, then the valuation produced by the high-end forest is used directly as the model's valuation. Otherwise, a weighted average of the valuations produced by the core forest and the high-end forest is used, where the weight of the core forest valuation is based upon nearness of the core model valuation to the 97.5 percentile selling price, while the weight of the high-end forest valuation is based on the nearness of the core forest valuation to the 99 percentile selling price.

In some embodiments, the facility uses valuations produced by the facility over time to calculate a price index for homes in particular geographic areas, which may be larger than, smaller than, or the same as the geographic areas that are the basis for individual valuation models. In order to calculate the index, the facility averages the valuations produced by the facility for houses in a geographic area at each a first and a second date, and generates an indication of the extent and direction of change. For example, the extent may be expressed in terms of dollars or some multiple of a particular number of dollars—such as $1,000—or as a percentage based upon either the first average valuation or the second valuation. The direction may be indicated by a plus or minus sign, an up or down arrow, etc. In some embodiments, the facility displays a visual indication of this price index as part of a visual representation of the corresponding geographic area, such as a map or an aerial photograph. Any visual representation that covers more than one geographic area may contain a price index indication for each such geographic area. In some embodiments, the facility provides these price indices in a variety of other ways, including a table, a chart, a data feed, etc.

It will be appreciated by those skilled in the art that the above-described facility may be straightforwardly adapted or extended in various ways. For example, the facility may use a wide variety of modeling techniques, house attributes, and/or data sources. The facility may display or otherwise present its valuations in a variety of ways. While the foregoing description makes reference to particular embodiments, the scope of the invention is defined solely by the claims that follow and the elements recited therein.

Claims

1. A computer-implemented method, in a computing system having a memory and a processor, for generating machine learning models to value homes located in a distinguished geographic area, comprising: retrieving, by the processor, home sales data for the distinguished geographic area, the home sales data comprising multiple entries, each entry indicating a selling price and a value for one or more home attributes;creating, by the processor, one or more machine learning classification trees by: for each distinguished classification tree of the one or more classification trees: selecting a subset of the multiple entries;selecting a subset of the one or more home attributes;for each of the selected home attributes, determining a range of values of the selected attribute among the selected entries;establishing a root node in the distinguished classification tree representing the range of values of each of the selected attributes; andfor each distinguished node of the tree that has not been identified as a leaf node, determining a greatest information gain resulting from one or more possible splits in the ranges of values represented by the distinguished node;when the greatest information gain exceeds an information gain identified for the distinguished node, establishing, for each of two subranges corresponding to the split with the greatest information gain, a child node of the distinguished node; andwhen the greatest information gain does not exceed the information gain identified for the distinguished node, identifying the distinguished node as a leaf node and calculating a mean selling price for homes represented by the leaf node.

2. The computer-implemented method of claim 1 further comprising: for each selected classification tree of at least one of the multiple classification trees:for each entry of one or more entries excluded from the selected entries for the selected classification tree: identifying a leaf node corresponding to the entry based on a match between one or more attribute values of the entry and one or more attribute ranges corresponding to the identified leaf node; anddetermining a difference between the mean selling price for homes represented by the identified leaf node and the selling price of the entry; andscoring the selected classification tree based on the determined one or more differences corresponding to each of the one or more entries.

3. The computer-implemented method of claim 2 further comprising: receiving attribute values for a distinguished home;identifying a certain leaf node in each of the multiple classification trees, wherein each certain leaf node is identified due to at least one attribute value for the distinguished home falling in one or more attribute ranges corresponding to the certain leaf node;determining the mean selling prices corresponding to each of the certain leaf nodes, wherein each mean selling price is weighted by the tree score for the classification tree containing the certain leaf node corresponding to that selling price;averaging the determined weighted mean selling prices; andreporting the average as an obtained valuation of the distinguished home.

4. The computer-implemented method of claim 3 further comprising: determining that a value for a particular attribute for the distinguished home is unavailable; andin response to the determination, imputing a value for the particular home attribute for the distinguished home.

5. The computer-implemented method of claim 4 further comprising: choosing, as the imputed value for the particular home attribute, a median value of the particular home attribute from among an identified set of homes sold in the distinguished geographic area.

6. The computer-implemented method of claim 3 further comprising: blending into the obtained valuation an earlier-reported valuation for the distinguished home by 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.

7. The computer-implemented method of claim 3 further comprising: blending into the obtained valuation an earlier-reported valuation for the distinguished home by generating a weighted average of the obtained valuation and the earlier-reported valuation in which the obtained valuation is more heavily weighted than the earlier-reported valuation.

8. A computer-readable medium, not constituting transitory signals, whose contents cause a computing system to perform a method for valuing homes located in a distinguished geographic area, the method comprising: receiving, over a computer network, home attributes for a distinguished home;obtaining, with a processor, a valuation for the distinguished home by applying, to the home attributes, a machine learning model trained at least in part by applying weights to portions of the model based on training items, each training item comprising attributes for a home in the distinguished geographic area and a selling price, wherein the valuation model includes: (1) a first component for all homes in the distinguished geographic area; and(2) a second component for a set of most highly-valued homes in the distinguished geographic area; andreporting the obtained valuation for the distinguished home.

9. The computer-readable medium of claim 8 wherein the valuation model is applied by first applying the first component for all homes in the distinguished geographic area, and using the obtained valuation to weight valuations generated for the home.

10. The computer-readable medium of claim 8, wherein the method further comprises: determining that a value for a particular home attribute for the distinguished home is unavailable; andin response to the determination, imputing a value for the particular home attribute for the distinguished home.

11. The computer-readable medium of claim 10, wherein the method further comprises choosing, as the imputed value for the particular home attribute, a median value of the particular home attribute from among an identified set of homes sold in the distinguished geographic area.

12. The computer-readable medium of claim 8, wherein the method further comprises blending into the obtained valuation an earlier-reported valuation for the distinguished home by 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.

13. The computer-readable medium of claim 8, wherein the method further comprises: using the model to produce first valuations for a group of homes in a portion of the distinguished geographic area at a first time index;determine a first average of the first valuations;use the model to produce second valuations for the group of homes in the portion of the distinguished geographic area at a second later time index;determine a second average of the second valuations; andgenerate an extent and direction of change between the first average and the second average.

14. The computer-readable medium of claim 13, wherein indications of the extent and direction of change are provided, in a graphical topological representation, in association with a portion of the graphical topological representation of the portion of the distinguished geographic area.

15. The computer-readable medium of claim 8, wherein the model is used to produce valuations for a group of homes in a portion of the distinguished geographic area; andwherein a map is provided to a user device with indications of the produced valuations, each valuation indication provided in association with graphical representations of a location of the home for which that valuation was produced.

16. One or more computer memories, not constituting transitory signals, collectively storing: a valuation model data structure, comprising multiple portions, capable of producing a valuation of a home when trained using sales information usable to value a home in a distinguished geographic area when applied to home attributes for a distinguished home, wherein the valuation model was trained using sales information for a set of homes in the distinguished geographic area, the sales information including home attributes and a selling price for each home in the set of homes, andwherein the training included weighting each portion of the portions of the valuation model based upon a level of success of the valuation model in valuing homes using the sales information.

17. The one or more computer memories of claim 16, further collectively storing instructions for a statistics module that, when executed by one or more processors, cause the one or more processors to: use the valuation model to produce first valuations for a group of homes in a portion of the distinguished geographic area at a first time index;determine a first average of the first valuations;use the valuation model to produce second valuations for the group of homes in the portion of the distinguished geographic area at a second later time index;determine a second average of the second valuations; andgenerate an extent and a direction of change between the first average and the second average.

18. The one or more computer memories of claim 17, wherein indications of the extent and the direction of change are provided, in a graphical topological representation, in association with a portion of the graphical topological representation that is for the portion of the distinguished geographic area.

19. The one or more computer memories of claim 16, wherein the valuation model is used to produce valuations for a group of homes in a portion of the distinguished geographic area; andwherein a map is provided to a client device with indications of the produced valuations, each valuation indication provided in association with graphical representations of a location of the home for which that valuation was produced.

20. The one or more computer memories of claim 16, wherein the stored valuation model data structure comprises: a first classification tree forest trained on sales information for a first group of homes in the distinguished geographic area; anda second classification tree forest trained on sales information for a second group of homes in the distinguished geographic area that is a proper subset of the first group of homes, wherein the second group of homes is selected based on the selling price for each home in the second group of homes being above a threshold for the distinguished geographic area.