{"id":348182,"date":"2018-10-29T05:24:18","date_gmt":"2018-10-29T12:24:18","guid":{"rendered":"http:\/\/www.esri.com\/arcgis-blog\/?post_type=blog&p=348182"},"modified":"2018-10-29T05:26:55","modified_gmt":"2018-10-29T12:26:55","slug":"using-forest-based-classification-and-regression-to-model-and-estimate-house-values","status":"publish","type":"blog","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/analytics\/using-forest-based-classification-and-regression-to-model-and-estimate-house-values","title":{"rendered":"Using Forest-based Classification and Regression to Model and Estimate House Values"},"author":7101,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","format":"standard","meta":{"_acf_changed":false,"_searchwp_excluded":""},"categories":[23341,37171,37141],"tags":[35661,34551,256992,25581],"industry":[],"product":[36561],"class_list":["post-348182","blog","type-blog","status-publish","format-standard","hentry","category-analytics","category-business","category-decision-support","tag-machine-learning","tag-prediction","tag-regression-classification","tag-spatial-statistics","product-arcgis-pro"],"acf":{"short_description":"ArcGIS Pro 2.2 has an exciting machine learning tool that can help make predictions, called Forest-based Classification and Regression. ","flexible_content":[{"acf_fc_layout":"content","content":"

The\u00a0ArcGIS Pro 2.2\u00a0release has an exciting new machine learning tool that can help make predictions. It\u2019s called\u00a0Forest-based Classification and Regression<\/em><\/strong><\/a>, and it lets analysts effectively design, test, and deploy predictive models.<\/p>\n"},{"acf_fc_layout":"image","image":{"ID":348192,"id":348192,"title":"0_0Q2NW8EQxy6uvjGd","filename":"0_0Q2NW8EQxy6uvjGd.png","filesize":18760,"url":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/analytics\/using-forest-based-classification-and-regression-to-model-and-estimate-house-values\/0_0q2nw8eqxy6uvjgd","alt":"","author":"7101","description":"","caption":"","name":"0_0q2nw8eqxy6uvjgd","status":"inherit","uploaded_to":348182,"date":"2018-10-27 20:06:58","modified":"2018-10-27 20:06:58","menu_order":0,"mime_type":"image\/png","type":"image","subtype":"png","icon":"https:\/\/www.esri.com\/arcgis-blog\/wp-includes\/images\/media\/default.png","width":489,"height":253,"sizes":{"thumbnail":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","thumbnail-width":213,"thumbnail-height":110,"medium":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","medium-width":464,"medium-height":240,"medium_large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","medium_large-width":489,"medium_large-height":253,"large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","large-width":489,"large-height":253,"1536x1536":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","1536x1536-width":489,"1536x1536-height":253,"2048x2048":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","2048x2048-width":489,"2048x2048-height":253,"card_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","card_image-width":489,"card_image-height":253,"wide_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/0_0Q2NW8EQxy6uvjGd.png","wide_image-width":489,"wide_image-height":253}},"image_position":"center","orientation":"horizontal","hyperlink":""},{"acf_fc_layout":"content","content":"

Forest-based Classification and Regression applies Leo Breiman\u2019s random forest algorithm, a popular supervised machine learning method used in classification and prediction. The tool allows analysts to easily incorporate tabular attributes, distance-based features, and explanatory rasters to build predictive models and expands predictive modeling to become accessible and possible for all GIS users.<\/p>\n

To show off what\u2019s possible with Forest-based Classification and Regression, we tackled a popular problem in the data science community: predicting home sale values. Let\u2019s take a look at a basic exercise to build a model that incorporates spatial factors to help improve the prediction of home sale prices in California.<\/p>\n

Predicting House Prices in California<\/strong><\/p>\n

We\u2019ll start by using the popular\u00a0California Housing Dataset<\/a>\u00a0from Kaggle, containing tracts in California with a series of aggregate attributes for the homes in each tract.<\/p>\n"},{"acf_fc_layout":"image","image":{"ID":348202,"id":348202,"title":"1_20HpNVtfiJNV4rc645UbDw","filename":"1_20HpNVtfiJNV4rc645UbDw.png","filesize":53778,"url":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/analytics\/using-forest-based-classification-and-regression-to-model-and-estimate-house-values\/1_20hpnvtfijnv4rc645ubdw","alt":"","author":"7101","description":"","caption":"","name":"1_20hpnvtfijnv4rc645ubdw","status":"inherit","uploaded_to":348182,"date":"2018-10-27 20:08:07","modified":"2018-10-27 20:08:07","menu_order":0,"mime_type":"image\/png","type":"image","subtype":"png","icon":"https:\/\/www.esri.com\/arcgis-blog\/wp-includes\/images\/media\/default.png","width":949,"height":353,"sizes":{"thumbnail":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","thumbnail-width":213,"thumbnail-height":79,"medium":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","medium-width":464,"medium-height":173,"medium_large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","medium_large-width":768,"medium_large-height":286,"large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","large-width":949,"large-height":353,"1536x1536":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","1536x1536-width":949,"1536x1536-height":353,"2048x2048":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","2048x2048-width":949,"2048x2048-height":353,"card_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","card_image-width":826,"card_image-height":307,"wide_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_20HpNVtfiJNV4rc645UbDw.png","wide_image-width":949,"wide_image-height":353}},"image_position":"center","orientation":"horizontal","hyperlink":""},{"acf_fc_layout":"content","content":"

It\u2019s tough to do anything meaningful just by looking at the above table, so let\u2019s make a map of each tract, symbolized by average home sale value at each location:<\/p>\n"},{"acf_fc_layout":"image","image":{"ID":348222,"id":348222,"title":"1_9KPepocP9Ux2vmKvMpGQNQ","filename":"1_9KPepocP9Ux2vmKvMpGQNQ.png","filesize":107866,"url":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/analytics\/using-forest-based-classification-and-regression-to-model-and-estimate-house-values\/1_9kpepocp9ux2vmkvmpgqnq","alt":"","author":"7101","description":"","caption":"","name":"1_9kpepocp9ux2vmkvmpgqnq","status":"inherit","uploaded_to":348182,"date":"2018-10-27 20:09:37","modified":"2018-10-27 20:09:37","menu_order":0,"mime_type":"image\/png","type":"image","subtype":"png","icon":"https:\/\/www.esri.com\/arcgis-blog\/wp-includes\/images\/media\/default.png","width":800,"height":711,"sizes":{"thumbnail":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","thumbnail-width":213,"thumbnail-height":189,"medium":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","medium-width":294,"medium-height":261,"medium_large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","medium_large-width":768,"medium_large-height":683,"large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","large-width":800,"large-height":711,"1536x1536":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","1536x1536-width":800,"1536x1536-height":711,"2048x2048":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","2048x2048-width":800,"2048x2048-height":711,"card_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","card_image-width":523,"card_image-height":465,"wide_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_9KPepocP9Ux2vmKvMpGQNQ.png","wide_image-width":800,"wide_image-height":711}},"image_position":"center","orientation":"horizontal","hyperlink":""},{"acf_fc_layout":"content","content":"

In this map, each point signifies the centroid of a tract in California. The color range represents the average home sale value of all homes in the tract. Blue represents low sales values, yellow represents medium sales value, and red represents the highest values.<\/p>\n

Just from viewing this map, do you notice any general pattern?<\/p>\n

You may notice that higher-priced homes are situated near the largest metropolitan areas. You may also notice that higher-priced homes are situated near the coastline. A quick exploratory chart in ArcGIS Pro helps us explore these patterns:<\/p>\n"},{"acf_fc_layout":"image","image":{"ID":348232,"id":348232,"title":"1_FVsK-lEOQGcVj7pl2YPULQ","filename":"1_FVsK-lEOQGcVj7pl2YPULQ.png","filesize":88734,"url":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/analytics\/using-forest-based-classification-and-regression-to-model-and-estimate-house-values\/1_fvsk-leoqgcvj7pl2ypulq","alt":"","author":"7101","description":"","caption":"","name":"1_fvsk-leoqgcvj7pl2ypulq","status":"inherit","uploaded_to":348182,"date":"2018-10-27 20:10:55","modified":"2018-10-27 20:10:55","menu_order":0,"mime_type":"image\/png","type":"image","subtype":"png","icon":"https:\/\/www.esri.com\/arcgis-blog\/wp-includes\/images\/media\/default.png","width":1000,"height":553,"sizes":{"thumbnail":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","thumbnail-width":213,"thumbnail-height":118,"medium":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","medium-width":464,"medium-height":257,"medium_large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","medium_large-width":768,"medium_large-height":425,"large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","large-width":1000,"large-height":553,"1536x1536":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","1536x1536-width":1000,"1536x1536-height":553,"2048x2048":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","2048x2048-width":1000,"2048x2048-height":553,"card_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","card_image-width":826,"card_image-height":457,"wide_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_FVsK-lEOQGcVj7pl2YPULQ.png","wide_image-width":1000,"wide_image-height":553}},"image_position":"center","orientation":"horizontal","hyperlink":""},{"acf_fc_layout":"content","content":"

Let\u2019s view the rest of the data in the provided table. Each record contains a few basic data points for all the homes in the tract:<\/p>\n"},{"acf_fc_layout":"image","image":{"ID":348242,"id":348242,"title":"1_YwsoevZw5c83Vu8ga-fB1w","filename":"1_YwsoevZw5c83Vu8ga-fB1w.png","filesize":111768,"url":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/analytics\/using-forest-based-classification-and-regression-to-model-and-estimate-house-values\/1_ywsoevzw5c83vu8ga-fb1w","alt":"","author":"7101","description":"","caption":"","name":"1_ywsoevzw5c83vu8ga-fb1w","status":"inherit","uploaded_to":348182,"date":"2018-10-27 20:11:34","modified":"2018-10-27 20:11:34","menu_order":0,"mime_type":"image\/png","type":"image","subtype":"png","icon":"https:\/\/www.esri.com\/arcgis-blog\/wp-includes\/images\/media\/default.png","width":1000,"height":428,"sizes":{"thumbnail":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","thumbnail-width":213,"thumbnail-height":91,"medium":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","medium-width":464,"medium-height":199,"medium_large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","medium_large-width":768,"medium_large-height":329,"large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","large-width":1000,"large-height":428,"1536x1536":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","1536x1536-width":1000,"1536x1536-height":428,"2048x2048":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","2048x2048-width":1000,"2048x2048-height":428,"card_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","card_image-width":826,"card_image-height":354,"wide_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2018\/10\/1_YwsoevZw5c83Vu8ga-fB1w.png","wide_image-width":1000,"wide_image-height":428}},"image_position":"center","orientation":"horizontal","hyperlink":""},{"acf_fc_layout":"content","content":"

The median house value for each tract is our variable to predict, and these attributes are likely important in helping estimate each value.<\/p>\n

We\u2019ll start by following the example provided by Aur\u00e9lien Geron in his book\u00a0Hands-On Machine Learning with Scikit-Learn and TensorFlow<\/em><\/a>, where\u00a0a random forest model<\/a>\u00a0was built using primarily non-spatial factors (i.e. the attributes in the table shown above). We\u2019ll compare this model to a second model where we start to bring in other GIS layers to assess how each tract\u2019s proximity to locations of interest may help the model improve when estimating median house values.<\/p>\n

Non-spatial Model<\/strong><\/p>\n

Our first model will follow the lead of the Hands-On Machine Learning with Scikit-Learn and TensorFlow example, using the following characteristics for each tract record:<\/p>\n