Valuation of homes using geographic regions of varying granularity
A facility for estimating a subject home's value is described. For each of one or more direct home attributes, the facility determines the value of the direct home attribute for the subject home. For each of a plurality of arbitrary geographic regions of different sizes containing the subject home, the facility determines information relating to the geographic region as a whole. The facility then subjects the determined values and information to a statistical home valuation model to obtain an estimated value of the subject home.
1. A method in a computing system for estimating a value of a subject home, comprising:
creating a training set comprising data of sale transactions of a plurality of homes selected from homes in two or more region sizes;
periodically training one or more machine learning models, comprising at least one of a gradient boosting machine, a support vector machine, or a neural network, using the created training set,
wherein independent variables of the one or more machine learning models comprise, for each of a plurality of region sizes:
an independent variable identifying a region of the region size containing the subject home,
one or more independent variables each identifying a neighboring region of the region size that borders the region of the region size containing the subject home, wherein, for a respective region size, the region containing the subject home is non-overlapping with the neighboring region, and
one or more independent variables each determined from an aggregation of values of a home attribute across all homes, in the region of the region size containing the subject home, for which a value of the home attribute is available;
generating, for each region corresponding to each of the plurality of region sizes and based on a latitude/longitude pair associated with the subject home, a single computer-readable geohash encoding value,
wherein the single computer-readable geohash encoding value of a first region size is generated by discarding a least significant digit of the single computer-readable geohash encoding value of a second region size, the first region size being at a next-less-granular level than the second region size;
determining, based on the single computer-readable geohash encoding value, values of the trained one or more machine learning model's independent variables for the subject home;
applying the trained one or more machine learning models to the determined independent variable values to produce an estimate of the value of the subject home.
2. The method of claim 1 , further comprising:
measuring a difference between the produced estimated value of the subject home and a selling price of the subject home.
3. The method of claim 1 , wherein the two or more region sizes are of different sizes.
4. The method of claim 1 , further comprising, for each region, defining the region to include an area specified for the region's region size.
5. The method of claim 1 , further comprising, for each region, defining the region to include an area within a percentage of a target area specified for the region's region size.
6. The method of claim 1 , further comprising, for each region, defining the region to include a number of homes specified for the region's region size.
7. The method of claim 1 , further comprising, for each region, defining the region to include a percentage of a number of homes specified for the region's region size.
8. One or more instances of non-transitory computer-readable media having contents configured to cause a computing system to perform a method for estimating a value of a subject home, the method comprising:
creating a training set comprising data of sale transactions of a plurality of homes selected from homes in two or more region sizes;
periodically training one or more machine learning models, comprising at least one of a gradient boosting machine, a support vector machine, or a neural network, using the created training set,
wherein independent variables of the one or more machine learning models comprise, for each of a plurality of region sizes:
an independent variable identifying a region of the region size containing the subject home,
one or more independent variables each identifying a neighboring region of the region size that borders the region of the region size containing the subject home, wherein, for a respective region size, the region containing the subject home is non-overlapping with the neighboring region, and
one or more independent variables each determined from an aggregation of values of a home attribute across all homes, in the region of the region size containing the subject home, for which a value of the home attribute is available;
generating, for each region corresponding to each of the plurality of region sizes and based on a latitude/longitude pair associated with the subject home, a single computer-readable geohash encoding value,
wherein the single computer-readable geohash encoding value of a first region size is generated by discarding a least significant digit of the single computer-readable geohash encoding value of a second region size, the first region size being at a next-less-granular level than the second region size:
determining, based on the single computer-readable geohash encoding value, values of the trained one or more machine learning model's independent variables for the subject home;
applying the trained one or more machine learning models to the determined independent variable values to produce an estimate of the value of the subject home.
9. The one or more instances of non-transitory computer-readable media of claim 8 , the method further comprising:
measuring a difference between the produced estimated value of the subject home and a selling price of the subject home.
10. The one or more instances of non-transitory computer-readable media of claim 8 wherein the two or more region sizes are of different sizes.
11. The one or more instances of non-transitory computer-readable media of claim 8 wherein the method further comprises, for each region, defining the region to include an area specified for the region's region size.
12. The one or more instances of non-transitory computer-readable media of claim 8 wherein the method further comprises, for each region, defining the region to include an area within a percentage of a target area specified for the region's region size.
13. The one or more instances of non-transitory computer-readable media of claim 8 wherein the method further comprises, for each region, defining the region to include a number of homes specified for the region's region size.
14. The one or more instances of non-transitory computer-readable media of claim 8 wherein the method further comprises, for each region, defining the region to include a percentage of a number of homes specified for the region's region size.
15. A computing system for estimating a value of a subject home, comprising:
at least one processor; and
a memory having contents configured to cause the at least one processor to perform a method, the method comprising:
creating a training set comprising data of sale transactions of a plurality of homes selected from homes in two or more region sizes;
periodically training one or more machine learning models, comprising at least one of a gradient boosting machine, a support vector machine, or a neural network, using the created training set,
wherein independent variables of the one or more machine learning models comprise, for each of a plurality of region sizes:
an independent variable identifying a region of the region size containing the subject home,
one or more independent variables each identifying a neighboring region of the region size that borders the region of the region size containing the subject home, wherein, for a respective region size, the region containing the subject home is non-overlapping with the neighboring region, and
one or more independent variables each determined from an aggregation of values of a home attribute across all homes, in the region of the region size containing the subject home, for which a value of the home attribute is available;
generating, for each region corresponding to each of the plurality of region sizes and based on a latitude/longitude pair associated with the subject home, a single computer-readable geohash encoding value,
wherein the single computer-readable geohash encoding value of a first region size is generated by discarding a least significant digit of the single computer-readable geohash encoding value of a second region size, the first region size being at a next-less-granular level than the second region size;
determining, based on the single computer-readable geohash encoding value, values of the trained one or more machine learning model's independent variables for the subject home;
applying the trained one or more machine learning models to the determined independent variable values to produce an estimate of the value of the subject home.
16. The computing system of claim 15 wherein the two or more region sizes are of different sizes.
17. The computing system of claim 15 wherein the method further comprises, for each region, defining the region to include a number of homes specified for the region's region size.