Analysing the factors of growth and spatial distribution of Airbnb properties across New York City

pip install pillow==9.0.0

pip install seaborn

pip install scikit-learn

%matplotlib inline
import matplotlib.pyplot as plt


from datetime import datetime as dt
import pandas as pd
import numpy as np
from IPython.display import display, HTML
from IPython.core.pylabtools import figsize
import seaborn as sns


# Machine Learning models
from sklearn.linear_model import LinearRegression
from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
import sklearn.metrics as metrics
from sklearn import preprocessing

# Arcgis api imports
import arcgis
from arcgis.geoenrichment import Country
from arcgis.features import summarize_data
from arcgis.features.enrich_data import enrich_layer
from arcgis.features import use_proximity 
from arcgis.gis import GIS
from arcgis.features import summarize_data

gis = GIS(profile='your_online_profile')

# Accessing NYCTracts
nyc_tract_full = gis.content.search('NYCTractData owner:api_data_owner', 'feature layer')[0]
nyc_tract_full

nyc_tracts_layer = nyc_tract_full.layers[0]

# Accessing airbnb NYC
airbnb_nyc2019 = gis.content.search('AnBNYC2019 owner:api_data_owner', 'feature layer')[0]
airbnb_nyc2019

airbnb_layer = airbnb_nyc2019.layers[0]

# NYC Tracts
m1 = gis.map('New York City')
m1.content.add(nyc_tracts_layer)
m1

# NYC Airbnb Properties
m = gis.map('Springfield Gardens, NY')
m.content.add(airbnb_layer)
m

# extracting the dataframe from the layer and visualize it as a pandas dataframe
pd.set_option('display.max_columns', 110)
sdf_airbnb_layer = pd.DataFrame.spatial.from_layer(airbnb_layer)
sdf_airbnb_layer.head(2)

agg_result = summarize_data.aggregate_points(point_layer=airbnb_layer,
                                             polygon_layer=nyc_tracts_layer,
                                             output_name='airbnb_counts'+ str(dt.now().microsecond))

{"cost": 50.968}

agg_result

# mapping the aggregated airbnb data with darker areas showing more airbnb properties per tract
aggr_map = gis.map('New York, NY')
aggr_map.content.add(agg_result)
aggr_map

agg_result_renderer = aggr_map.content.renderer(0)
agg_result_renderer.smart_mapping().class_breaks_renderer(break_type="color", field="Point_Cloud")

airbnb_count_by_tract = agg_result.layers[0]

sdf_airbnb_count_by_tract = airbnb_count_by_tract.query().sdf

sdf_airbnb_count_by_tract = sdf_airbnb_count_by_tract.sort_values('geoid')
sdf_airbnb_count_by_tract.head()

# Displaying the various data topic available for geoenrichment for USA in the Esri database
usa = Country.get('US')
type(usa)
usa_data = usa.data_collections
df_usa_data = pd.DataFrame(usa_data)
df_usa_data.head()

# Filtering the unique topic under dataCollectionID
df_usa_data.reset_index(inplace=True)
list(df_usa_data.dataCollectionID.unique())

['1yearincrements',
 '5yearincrements',
 'Age',
 'agebyracebysex',
 'agebyracebysex2010',
 'agebyracebysex2020',
 'AgeDependency',
 'AtRisk',
 'AutomobilesAutomotiveProducts',
 'BabyProductsToysGames',
 'basicFactsForMobileApps',
 'businesses',
 'CivicActivitiesPoliticalAffiliation',
 'classofworker',
 'clothing',
 'ClothingShoesAccessories',
 'commute',
 'crime',
 'DaytimePopulation',
 'disability',
 'disposableincome',
 'DniRates',
 'education',
 'educationalattainment',
 'ElectronicsInternet',
 'employees',
 'EmploymentUnemployment',
 'entertainment',
 'financial',
 'FinancialInsurance',
 'food',
 'foodstampsSNAP',
 'gender',
 'Generations',
 'GroceryAlcoholicBeverages',
 'groupquarters',
 'Health',
 'healthinsurancecoverage',
 'HealthPersonalCare',
 'HealthPersonalCareCEX',
 'heatingfuel',
 'hispanicorigin',
 'HistoricalHouseholds',
 'HistoricalHousing',
 'HistoricalPopulation',
 'HomeImprovementGardenLawn',
 'homevalue',
 'HouseholdGoodsFurnitureAppliances',
 'householdincome',
 'households',
 'householdsbyageofhouseholder',
 'HouseholdsByIncome',
 'householdsbyraceofhouseholder',
 'householdsbysize',
 'householdtotals',
 'householdtype',
 'housingbyageofhouseholder',
 'housingbyraceofhouseholder',
 'housingbysize',
 'housingcosts',
 'HousingHousehold',
 'housingunittotals',
 'incomebyage',
 'industry',
 'industrybynaicscode',
 'industrybysiccode',
 'InternetComputerUsage',
 'KeyGlobalFacts',
 'KeyUSFacts',
 'language',
 'LeisureActivitiesLifestyle',
 'LifeInsurancePensions',
 'lifemodegroupsNEW',
 'maritalstatustotals',
 'MediaMagazinesNewspapers',
 'MediaRadioOtherAudio',
 'MediaTVViewing',
 'miscellaneous',
 'networth',
 'NonHispanicOrigin',
 'occupation',
 'OwnerRenter',
 'PetsPetProducts',
 'PhonesYellowPages',
 'Policy',
 'population',
 'populationtotals',
 'presenceofchildren',
 'PsychographicsAdvertising',
 'PsychographicsFood',
 'PsychographicsLifestyle',
 'PsychographicsMedia',
 'PsychographicsShopping',
 'RaceAndEthnicity',
 'raceandhispanicorigin',
 'restaurants',
 'RetailDemandbyNAICS',
 'sales',
 'schoolenrollment',
 'shopping',
 'spendingFactsForMobileApps',
 'SpendingTotal',
 'sports',
 'tapestryadultsNEW',
 'tapestryhouseholdsNEW',
 'TapestryNEW',
 'transportation',
 'TravelCEX',
 'travelMPI',
 'unitsinstructure',
 'urbanizationgroupsNEW',
 'vacant',
 'vehiclesavailable',
 'veterans',
 'Wealth',
 'women',
 'yearbuilt',
 'yearmovedin']

df_usa_data[df_usa_data['alias'].str.contains('Nonprofit')]                        

enrichment_variables = {'classofworker.ACSCIVEMP':      'Employed Population Age 16+',
 'classofworker.ACSMCIVEMP':                      'Employed Male Pop Age 16+',
 'classofworker.ACSMPRIVNP':                      'Male 16+Priv Nonprofit',
 'classofworker.ACSMEPRIVP':                      'Male 16+:Priv Profit Empl',
 'classofworker.ACSMSELFI':                       'Male 16+:Priv Profit Self Empl',
 'classofworker.ACSMSTGOV':                       'Male 16+:State Govt Wrkr',
 'classofworker.ACSMFEDGOV':                      'Male 16+:Fed Govt Wrkr',
 'classofworker.ACSMSELFNI':                      'Male 16+:Self-Emp Not Inc',
 'classofworker.ACSMUNPDFM':                      'Male 16+:Unpaid Family Wrkr',              
 'classofworker.ACSFCIVEMP':                      'Female Pop Age 16+',
 'classofworker.ACSFEPRIVP':                      'Female 16+:Priv Profit Empl',
 'classofworker.ACSFSELFI':                       'Female 16+:Priv Profit Self Empl',                      
 'classofworker.ACSFPRIVNP':                      'Female 16+:Priv Nonprofit',
 'classofworker.ACSFLOCGOV':                      'Female 16+:Local Govt Wrkr',
 'classofworker.ACSFSTGOV':                       'Female 16+:State Govt Wrkr',
 'classofworker.ACSFFEDGOV':                      'Female 16+:Fed Govt Wrkr',                      
 'classofworker.ACSFSELFNI':                      'Female 16+:Self-Emp Not Inc',                      
 'classofworker.ACSFUNPDFM':                      'Female 16+:Unpaid Family Wrkr',                      
 'gender.MEDAGE_CY':                              '2019 Median Age',
 'Generations.GENALPHACY':                        '2019 Generation Alpha Population',
 'Generations.GENZ_CY':                           '2019 Generation Z Population',
 'Generations.MILLENN_CY':                        '2019 Millennial Population',
 'Generations.GENX_CY':                           '2019 Generation X Population',
 'Generations.BABYBOOMCY':                        '2019 Baby Boomer Population',
 'Generations.OLDRGENSCY':                        '2019 Silent & Greatest Generations Population',
 'Generations.GENBASE_CY':                        '2019 Population by Generation Base',
 'populationtotals.POPDENS_CY':                   '2019 Population Density',
 'DaytimePopulation.DPOP_CY':                     '2019 Total Daytime Population',
 'raceandhispanicorigin.WHITE_CY':                '2019 White Population',
 'raceandhispanicorigin.BLACK_CY':                '2019 Black Population',
 'raceandhispanicorigin.AMERIND_CY':              '2019 American Indian Population',
 'raceandhispanicorigin.ASIAN_CY':                '2019 Asian Population',
 'raceandhispanicorigin.PACIFIC_CY':              '2019 Pacific Islander Population',
 'raceandhispanicorigin.OTHRACE_CY':              '2019 Other Race Population',
 'raceandhispanicorigin.DIVINDX_CY':              '2019 Diversity Index',
 'households.ACSHHBPOV':                          'HHs: Inc Below Poverty Level',
 'households.ACSHHAPOV':                          'HHs:Inc at/Above Poverty Level',
 'households.ACSFAMHH':                           'ACS Family Households',
 'businesses.S01_BUS':                            'Total Businesses (SIC)',
 'businesses.N05_BUS':                            'Construction Businesses (NAICS)',
 'businesses.N08_BUS':                            'Retail Trade Businesses (NAICS)',
 'businesses.N21_BUS':                            'Transportation/Warehouse Bus (NAICS)',
 'ElectronicsInternet.MP09147a_B':                'Own any tablet',
 'ElectronicsInternet.MP09148a_B':                'Own any e-reader',
 'ElectronicsInternet.MP19001a_B':                'Have access to Internet at home',                
 'ElectronicsInternet.MP19070a_I':                'Index: Spend 0.5-0.9 hrs online(excl email/IM .',               
 'ElectronicsInternet.MP19071a_B':                'Spend <0.5 hrs online (excl email/IM time) daily',
 'populationtotals.TOTPOP_CY':                    '2019 Total Population',              
 'gender.MALES_CY':                               '2019 Male Population',
 'gender.FEMALES_CY':                             '2019 Female Population',
 'industry.EMP_CY':                               '2019 Employed Civilian Pop 16+',
 'industry.UNEMP_CY':                             '2019 Unemployed Population 16+',                     
 'industry.UNEMPRT_CY':                           '2019 Unemployment Rate',
 'commute.ACSWORKERS':                            'ACS Workers Age 16+',
 'commute.ACSDRALONE':                            'ACS Workers 16+: Drove Alone',
 'commute.ACSCARPOOL':                            'ACS Workers 16+: Carpooled',
 'commute.ACSPUBTRAN':                            'ACS Workers 16+: Public Transportation',
 'commute.ACSBUS':                                'ACS Workers 16+: Bus',
 'commute.ACSSTRTCAR':                            'ACS Workers 16+: Streetcar',
 'commute.ACSSUBWAY':                             'ACS Workers 16+: Subway',
 'commute.ACSRAILRD':                             'ACS Workers 16+: Railroad',
 'commute.ACSFERRY':                              'ACS Workers 16+: Ferryboat',
 'commute.ACSTAXICAB':                            'ACS Workers 16+: Taxicab',           
 'commute.ACSMCYCLE':                             'ACS Workers 16+: Motorcycle',
 'commute.ACSBICYCLE':                            'ACS Workers 16+: Bicycle',                             
 'commute.ACSWALKED':                             'ACS Workers 16+: Walked',
 'commute.ACSOTHTRAN':                            'ACS Workers 16+: Other Means',
 'commute.ACSWRKHOME':                            'ACS Wrkrs 16+: Worked at Home',
 'OwnerRenter.OWNER_CY':                          '2019 Owner Occupied HUs', 
 'OwnerRenter.RENTER_CY':                         '2019 Renter Occupied HUs', 
 'vacant.VACANT_CY':                              '2019 Vacant Housing Units', 
 'homevalue.MEDVAL_CY':                           '2019 Median Home Value',
 'housingunittotals.TOTHU_CY':                    '2019 Total Housing Units',
 'yearbuilt.ACSMEDYBLT':                          'ACS Median Year Structure Built: HUs',
 'SpendingTotal.X1001_X':                         '2019 Annual Budget Exp',
 'transportation.X6001_X':                        '2019 Transportation',
 'households.ACSTOTHH':                           'ACS Total Households',
 'DaytimePopulation.DPOPWRK_CY':                  '2019 Daytime Pop: Workers',
 'DaytimePopulation.DPOPRES_CY':                  '2019 Daytime Pop: Residents',
 'DaytimePopulation.DPOPDENSCY':                  '2019 Daytime Pop Density',
 'occupation.OCCPROT_CY':                         '2019 Occupation: Protective Service',
 'occupation.OCCFOOD_CY':                         '2019 Occupation: Food Preperation',
 'occupation.OCCPERS_CY':                         '2019 Occupation: Personal Care',
 'occupation.OCCADMN_CY':                         '2019 Occupation: Office/Admin',
 'occupation.OCCCONS_CY':                         '2019 Occupation: Construction/Extraction',
 'occupation.OCCPROD_CY':                         '2019 Occupation: Production'
                  }

# Enrichment operation using ArcGIS API for Python 
enrichment_variables_df = pd.DataFrame.from_dict(enrichment_variables, orient='index',columns=['Variable Definition'])
enrichment_variables_df.reset_index(level=0, inplace=True)
enrichment_variables_df.columns = ['AnalysisVariable','Variable Definition']
enrichment_variables_df.head()

# Convertng the variables names to list for passing them to the enrichment tool
variable_names = enrichment_variables_df['AnalysisVariable'].tolist()

# checking the firt few values of the list
variable_names[1:5]

['classofworker.ACSMCIVEMP',
 'classofworker.ACSMPRIVNP',
 'classofworker.ACSMEPRIVP',
 'classofworker.ACSMSELFI']

# Data Enriching operation
airbnb_count_by_tract_enriched = enrich_layer(airbnb_count_by_tract,
                                              analysis_variables = variable_names,
                                              output_name='airbnb_tract_enrich1'+ str(dt.now().microsecond))

{"messageCode": "AO_100047", "message": "Enrichment may not be available for some features."}
{"messageCode": "AO_100288", "message": "Unable to detect the country for one or more features."}
{"messageCode": "AO_100047", "message": "Enrichment may not be available for some features."}
{"messageCode": "AO_100000", "message": "Variables [commute.ACSRAILRD] are not defined for country 'US'."}
{"messageCode": "AO_100000", "message": "Variables [households.ACSFAMHH, households.ACSTOTHH] are not defined for country 'US'."}
{"cost": -1}

# Extracting the resulting enriched dataframe after the geoenrichment method
sdf_airbnb_count_by_tract_enriched = airbnb_count_by_tract_enriched.layers[0].query().sdf

# Visualizing the data as a pandas dataframe
print(sdf_airbnb_count_by_tract_enriched.columns)
sdf_airbnb_count_by_tract_enriched_sorted = sdf_airbnb_count_by_tract_enriched.sort_values('geoid')
sdf_airbnb_count_by_tract_enriched_sorted.head()

Index(['OBJECTID', 'statefp', 'countyfp', 'tractce', 'geoid', 'name',
       'namelsad', 'mtfcc', 'funcstat', 'aland',
       ...
       'DPOPWRK_CY', 'DPOPRES_CY', 'DPOPDENSCY', 'OCCPROT_CY', 'OCCFOOD_CY',
       'OCCPERS_CY', 'OCCADMN_CY', 'OCCCONS_CY', 'OCCPROD_CY', 'SHAPE'],
      dtype='object', length=106)

enrichment_variables_df.head()

enrichment_variables_copy = enrichment_variables_df.copy()
enrichment_variables_copy.head(2)

enrichment_variables_copy['AnalysisVariable'] = enrichment_variables_copy.AnalysisVariable.str.split(pat='.', expand=True)[1]
enrichment_variables_copy

enrichment_variables_copy.set_index("AnalysisVariable", drop=True, inplace=True)
dictionary = enrichment_variables_copy.to_dict()
new_columns = dictionary['Variable Definition']

# Field renamed and new dataframe visualized
pd.set_option('display.max_columns', 150)
sdf_airbnb_count_by_tract_enriched_sorted.rename(columns=new_columns, inplace=True)
sdf_airbnb_count_by_tract_enriched_sorted.head()

gis.content.search('NYCBusStop owner:api_data_owner')

[]

busi_distr = gis.content.search('BusinessDistricts owner:api_data_owner', 'feature layer')[0]
cbd = gis.content.search('NYCBD owner:api_data_owner', 'feature layer')[0]
bus_stop = gis.content.search('NYCBusStop owner:api_data_owner', 'feature layer')[0]
hotels = gis.content.search('NYCHotels owner:api_data_owner', 'feature layer')[0]
railroad = gis.content.search('NYCRailroad owner:api_data_owner', 'feature layer')[0]
subwy_rt = gis.content.search('NYCSubwayRoutes owner:api_data_owner', 'feature layer')[0]
subwy_stn = gis.content.search('NYCSubwayStation owner:api_data_owner', 'feature layer')[0]

# accessing the various city feature shapefile from arcgis portal
busi_distr, cbd, bus_stop, hotels, railroad, subwy_rt, subwy_stn 

(<Item title:"tract_busi_distrs_dist874392" type:Feature Layer Collection owner:api_data_owner>,
 <Item title:"NYCBD" type:Feature Layer Collection owner:api_data_owner>,
 <Item title:"ny_tract_bus_stop_dist452536" type:Feature Layer Collection owner:api_data_owner>,
 <Item title:"ny_tract_hotel_dist128626" type:Feature Layer Collection owner:api_data_owner>,
 <Item title:"tract_railroad_dist613260" type:Feature Layer Collection owner:api_data_owner>,
 <Item title:"ny_tract_subway_routes_dist709262" type:Feature Layer Collection owner:api_data_owner>,
 <Item title:"ny_tract_subway_station_dist691504" type:Feature Layer Collection owner:api_data_owner>)

bus_stop_lyr = bus_stop.layers[0]
cbd_lyr = cbd.layers[0] 
hotels_lyr = hotels.layers[0] 
subwy_stn_lyr =subwy_stn.layers[0]
subwy_rt_lyr = subwy_rt.layers[0] 
railroad_lyr = railroad.layers[0]
busi_distrs_lyr = busi_distr.layers[0] 

# Avoid warning for chain operation
pd.set_option('mode.chained_assignment', None) 

# Estimating Tract to hotel distances
tract_hotel_dist = use_proximity.find_nearest(nyc_tracts_layer,
                                              hotels_lyr,
                                              measurement_type='StraightLine',
                                              max_count=1,
                                              output_name='ny_tract_hotel_dist1' + str(dt.now().microsecond))

{"cost": 2.555}

tract_hotel_dist.layers

[<FeatureLayer url:"https://services7.arcgis.com/JEwYeAy2cc8qOe3o/arcgis/rest/services/ny_tract_hotel_dist1396959/FeatureServer/0">,
 <FeatureLayer url:"https://services7.arcgis.com/JEwYeAy2cc8qOe3o/arcgis/rest/services/ny_tract_hotel_dist1396959/FeatureServer/1">]

tract_hotel_dist_lyr = tract_hotel_dist.layers[1]
sdf_tract_hotel_dist_lyr = pd.DataFrame.spatial.from_layer(tract_hotel_dist_lyr)
sdf_tract_hotel_dist_lyr.head()

# Final hotel Distances in feet — Here in each row column "hotel_dist" returns the distance of the nearest hotel from that tract indicated by its geoids.
# For example in the first row the tract with ID 36005000100 has a nearest hotel at 5571.75 feet away from it. 
sdf_tract_hotel_dist_lyr_new = sdf_tract_hotel_dist_lyr[['From_geoid', 'Total_Miles']]

# 1 mile = 5280 feet
sdf_tract_hotel_dist_lyr_new['hotel_dist'] = round(sdf_tract_hotel_dist_lyr_new['Total_Miles'] * 5280, 2)
sdf_tract_hotel_dist_lyr_new.sort_values('From_geoid').head()

# Estimating Busstop distances from tracts
tract_bustop_dist = use_proximity.find_nearest(nyc_tracts_layer,
                                               bus_stop_lyr,
                                               measurement_type='StraightLine',
                                               max_count=1,
                                               output_name='ny_tract_bus_stop_dist'+ str(dt.now().microsecond))
tract_bustop_dist_lyr = tract_bustop_dist.layers[1]
sdf_tract_bustop_dist_lyr =tract_bustop_dist_lyr.query().sdf

{"cost": 3.846}

sdf_tract_bustop_dist_lyr

# Final Bustop Distances in feet — Here in each row column "busstop_dist" returns the distance of the nearest bus stop 
# from that tract indicated by its geoids 
sdf_tract_bustop_dist_lyr_new = sdf_tract_bustop_dist_lyr[['From_geoid', 'Total_Miles']]
sdf_tract_bustop_dist_lyr_new['busstop_dist'] = round(sdf_tract_bustop_dist_lyr_new['Total_Miles'] * 5280, 2)
sdf_tract_bustop_dist_lyr_new.sort_values('From_geoid').head()

# estimating number of bus stops per tract
num_bustops_tracts = summarize_data.aggregate_points(point_layer=bus_stop_lyr,
                                                   polygon_layer=nyc_tracts_layer,
                                                   output_name='bustops_by_tracts'+ str(dt.now().microsecond)) 

{"cost": 3.846}

num_bustops_tracts_lyr = num_bustops_tracts.layers[0]
sdf_num_bustops_tracts_lyr = pd.DataFrame.spatial.from_layer(num_bustops_tracts_lyr)
sdf_num_bustops_tracts_lyr.head()

# Number of Bus stops per tract — Here in each row column "num_bustop" returns the number of bus stops inside respective tracts 
sdf_num_bustops_tracts_lyr_new = sdf_num_bustops_tracts_lyr[['geoid', 'Point_Count']] 
sdf_num_bustops_tracts_lyr_new = sdf_num_bustops_tracts_lyr_new.rename(columns={'Point_Count':'num_bustop'})
sdf_num_bustops_tracts_lyr_new.sort_values('geoid').head()

# estimating tracts distances from CBD 
tract_cbd_dist=use_proximity.find_nearest(nyc_tracts_layer,
                                          cbd_lyr,
                                          measurement_type='StraightLine',
                                          max_count=1,
                                          output_name='ny_tract_cbd_dist'+ str(dt.now().microsecond))
tract_cbd_dist_lyr = tract_cbd_dist.layers[1]
sdf_tract_cbd_dist_lyr = tract_cbd_dist_lyr.query().sdf
sdf_tract_cbd_dist_lyr.head()

{"cost": 2.168}

# Final CBD distances in feet — Here in each row the column "cbd_dst" returns the distance of the CBD from respective tracts
sdf_tract_cbd_dist_lyr_new = sdf_tract_cbd_dist_lyr[['From_geoid', 'Total_Miles']]
sdf_tract_cbd_dist_lyr_new['cbd_dist'] = round(sdf_tract_cbd_dist_lyr_new['Total_Miles'] * 5280, 2) 
sdf_tract_cbd_dist_lyr_new.sort_values('From_geoid').head()

# Estimating NYCSubwayStation distances from tracts 
tract_subwy_stn_dist = use_proximity.find_nearest(nyc_tracts_layer,
                                                  subwy_stn_lyr,
                                                  measurement_type='StraightLine',
                                                  max_count=1,
                                                  output_name='ny_tract_subway_station_dist'+ str(dt.now().microsecond))
tract_subwy_stn_dist_lyr = tract_subwy_stn_dist.layers[1]
sdf_tract_subwy_stn_dist_lyr = pd.DataFrame.spatial.from_layer(tract_subwy_stn_dist_lyr)
sdf_tract_subwy_stn_dist_lyr.head()

{"cost": 2.599}

# Final Tract to NYC Subway Station distances in feet — Here in each row, column "subwy_stn_dist" returns the distance of
# the nearest subway station from that tract
sdf_tract_subwy_stn_dist_lyr_new = sdf_tract_subwy_stn_dist_lyr[['From_geoid', 'Total_Miles']]
sdf_tract_subwy_stn_dist_lyr_new['subwy_stn_dist'] = round(sdf_tract_subwy_stn_dist_lyr_new['Total_Miles'] * 5280, 2) 
sdf_tract_subwy_stn_dist_lyr_new.sort_values('From_geoid').head()

# Estimating distances to NYCSubwayRoutes
tract_subwy_rt_dist=use_proximity.find_nearest(nyc_tracts_layer,
                                               subwy_rt_lyr,
                                               measurement_type='StraightLine',
                                               max_count=1,
                                               output_name='ny_tract_subway_routes_dist'+ str(dt.now().microsecond))
tract_subwy_rt_dist_lyr = tract_subwy_rt_dist.layers[1]
sdf_tract_subwy_rt_dist_lyr = tract_subwy_rt_dist_lyr.query().sdf
sdf_tract_subwy_rt_dist_lyr.head()

{"cost": 2.191}

# Final Tract to NYCSubwayRoutes distances in feet — Here in each row, column "subwy_rt_dist" returns the distance of
# the nearest subway route from that tract
sdf_tract_subwy_rt_dist_lyr_new = sdf_tract_subwy_rt_dist_lyr[['From_geoid', 'Total_Miles']]
sdf_tract_subwy_rt_dist_lyr_new['subwy_rt_dist'] = round(sdf_tract_subwy_rt_dist_lyr_new['Total_Miles'] * 5280, 2) 
sdf_tract_subwy_rt_dist_lyr_new.sort_values('From_geoid').head()

# Estimating distances to NYCRailroad
tract_railroad_dist = use_proximity.find_nearest(nyc_tracts_layer,
                                           railroad_lyr,
                                           measurement_type='StraightLine',
                                           max_count=1,
                                           output_name='tract_railroad_dist'+ str(dt.now().microsecond))
tract_railroad_dist_lyr = tract_railroad_dist.layers[1]
sdf_tract_railroad_dist_lyr = pd.DataFrame.spatial.from_layer(tract_railroad_dist_lyr)
sdf_tract_railroad_dist_lyr.head()

{"cost": 2.168}

# Final Tract to NYCRailroad distances in feet — Here in each row, column "railroad_dist" returns the distance of
# the nearest rail road route from that tract
sdf_tract_railroad_dist_lyr_new = sdf_tract_railroad_dist_lyr[['From_geoid', 'Total_Miles']]
sdf_tract_railroad_dist_lyr_new['railroad_dist'] = round(sdf_tract_railroad_dist_lyr_new['Total_Miles'] * 5280, 2) 
sdf_tract_railroad_dist_lyr_new.sort_values('From_geoid').head()

# Estimating distances to NYC Business Districts
tract_busi_distrs_dist = use_proximity.find_nearest(nyc_tracts_layer,
                                                      busi_distrs_lyr,
                                                      measurement_type='StraightLine',
                                                      max_count=1,
                                                      output_name='tract_busi_distrs_dist'+ str(dt.now().microsecond))
tract_busi_distrs_dist_lyr = tract_busi_distrs_dist.layers[1]
sdf_tract_busi_distrs_dist_lyr = pd.DataFrame.spatial.from_layer(tract_busi_distrs_dist_lyr)
sdf_tract_busi_distrs_dist_lyr.head()

{"cost": 2.241}

# Final Tract to NYC Businesss Districts distances in feet — Here in each row, column "busi_distr_dist" returns the distance of the CBD from respective tracts
sdf_tract_busi_distrs_dist_lyr_new = sdf_tract_busi_distrs_dist_lyr[['From_geoid', 'Total_Miles']]
sdf_tract_busi_distrs_dist_lyr_new['busi_distr_dist'] = round(sdf_tract_busi_distrs_dist_lyr_new['Total_Miles'] * 5280, 2) 
sdf_tract_busi_distrs_dist_lyr_new.sort_values('From_geoid').head()

# Name of the borough, inside which the tracts are located 
ny_tract_boro = gis.content.search('NYCTractBorough owner:api_data_owner', 'feature layer')[0]
ny_tract_boro_lyr = ny_tract_boro.layers[0]
sdf_ny_tract_boro_lyr = pd.DataFrame.spatial.from_layer(ny_tract_boro_lyr)
sdf_ny_tract_boro_lyr_new = sdf_ny_tract_boro_lyr[['geoid', 'boro_name']]
sdf_ny_tract_boro_lyr_new.sort_values('geoid').head()

tract_merge_dist = sdf_tract_hotel_dist_lyr_new.merge(sdf_tract_subwy_rt_dist_lyr_new,
                                                           on='From_geoid', suffixes=('_left1', '_right1')).merge(sdf_tract_railroad_dist_lyr_new,
                                                           on='From_geoid', suffixes=('_left2', '_right2')).merge(sdf_tract_subwy_stn_dist_lyr_new,
                                                           on='From_geoid', suffixes=('_left3', '_right3')).merge(sdf_tract_busi_distrs_dist_lyr_new,
                                                           on='From_geoid', suffixes=('_left4', '_right4')).merge(sdf_tract_cbd_dist_lyr_new, on='From_geoid')
tract_merge_dist_new = tract_merge_dist[['From_geoid',
                                         'hotel_dist',
                                         'subwy_rt_dist',
                                         'railroad_dist',
                                         'subwy_stn_dist',
                                         'busi_distr_dist',
                                         'cbd_dist']]
tract_merge_dist_new = tract_merge_dist_new.rename(columns={'From_geoid':'geoid'})
tract_merge_dist_new.sort_values('geoid').head()

# merging number of bus stop and borough name
tract_merge_dist_new = tract_merge_dist_new.merge(sdf_num_bustops_tracts_lyr_new,
                                                 on='geoid').merge(sdf_ny_tract_boro_lyr_new,
                                                 on='geoid') 
tract_merge_dist_new = tract_merge_dist_new.sort_values('geoid')
tract_merge_dist_new.head()

# Accessing the airbnb count for each tract
sdf_airbnb_count_by_tract_new = sdf_airbnb_count_by_tract[['geoid','Point_Count']]
sdf_airbnb_count_by_tract_new = sdf_airbnb_count_by_tract_new.rename(columns={'Point_Count':'total_airbnb'})
sdf_airbnb_count_by_tract_new.head()

# preparing the final distance table with airbnb count by tract
tract_merge_dist_all = sdf_airbnb_count_by_tract_new.merge(tract_merge_dist_new, on='geoid')
tract_merge_dist_all.head()

tract_merge_dist_all.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 2167 entries, 0 to 2166
Data columns (total 10 columns):
 #   Column           Non-Null Count  Dtype  
---  ------           --------------  -----  
 0   geoid            2167 non-null   string 
 1   total_airbnb     2167 non-null   Int32  
 2   hotel_dist       2167 non-null   Float64
 3   subwy_rt_dist    2167 non-null   Float64
 4   railroad_dist    2167 non-null   Float64
 5   subwy_stn_dist   2167 non-null   Float64
 6   busi_distr_dist  2167 non-null   Float64
 7   cbd_dist         2167 non-null   Float64
 8   num_bustop       2167 non-null   Int32  
 9   boro_name        2167 non-null   string 
dtypes: Float64(6), Int32(2), string(2)
memory usage: 169.4 KB

tract_merge_dist_final = pd.get_dummies(tract_merge_dist_all, columns=['boro_name'])
tract_merge_dist_final.head()

sdf_airbnb_count_by_tract_enriched_sorted_new = sdf_airbnb_count_by_tract_enriched_sorted.drop(['AnalysisArea',
                                                                                                'ENRICH_FID',
                                                                                                'HasData',
                                                                                                'ID',
                                                                                                'OBJECTID',
                                                                                                'Point_Count',
                                                                                                'SHAPE',                      
                                                                                                'aggregationMethod',
                                                                                                'aland',
                                                                                                'apportionmentConfidence',
                                                                                                'awater',
                                                                                                'countyfp',
                                                                                                'funcstat',
                                                                                                'intptlat',
                                                                                                'intptlon',
                                                                                                'mtfcc',
                                                                                                'name',
                                                                                                'namelsad',
                                                                                                'populationToPolygonSizeRating',
                                                                                                'sourceCountry',
                                                                                                'statefp','tractce'], axis=1)
sdf_airbnb_count_by_tract_enriched_sorted_new.shape

(2167, 84)

# checking the rows of the table for nan values
row_with_null = sdf_airbnb_count_by_tract_enriched_sorted_new.isnull().any(axis=1)

# printing the row which has nan values
sdf_airbnb_count_by_tract_enriched_sorted_new[row_with_null]

# checking total number of nan values
nan_test = sdf_airbnb_count_by_tract_enriched_sorted_new.drop(['geoid'], axis=1)
np.isnan(nan_test).sum().sum()

0

sdf_airbnb_count_by_tract_enriched_sorted_fill = sdf_airbnb_count_by_tract_enriched_sorted_new.fillna(0)

#nan rechecked
nan_test = sdf_airbnb_count_by_tract_enriched_sorted_fill.drop(['geoid'], axis=1)
np.isnan(nan_test).sum().sum()

0

final_df = pd.merge(tract_merge_dist_final,
                    sdf_airbnb_count_by_tract_enriched_sorted_fill,
                    left_on = 'geoid',
                    right_on = 'geoid',
                    how = 'left')

print(final_df.shape)
final_df.head()

(2167, 97)

# rechecking nan values of the final dataframe
final_nan_test = final_df.drop('geoid', axis=1)
np.isnan(final_nan_test).sum().sum()

0

# Creating feature data 
X = final_df.drop(['geoid','total_airbnb'], axis=1)

# Creating target data  -- the number airbnb per tract
y = pd.DataFrame(final_df['total_airbnb'])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.10, random_state = 20)

print(X_train.shape)
print(y_train.shape)

print(X_test.shape)
print(y_test.shape)

# Converting the target into 1d array
y_train_array = y_train.values.flatten()
y_test_array = y_test.values.flatten() 

print(y_train_array.shape)
print(y_test_array.shape)

(1950, 95)
(1950, 1)
(217, 95)
(217, 1)
(1950,)
(217,)

scaler = preprocessing.RobustScaler()

X_train_scaled = pd.DataFrame(scaler.fit_transform(X_train), columns = X_train.columns) 
X_test_scaled = pd.DataFrame(scaler.fit_transform(X_test), columns=X_test.columns)

# Random forest with scaled data
# for the best parameters a grid search could be done which could take some time
# however this model uses the default parameters of RF algorithm, while the estimators are changed till the best fit is obtained
model_RF = RandomForestRegressor(n_estimators = 500, random_state=43)

# Train the model
model_RF.fit(X_train_scaled, y_train_array)

# Training metrics for Random forest model
print('Training metrics for Random forest model using scaled data')
ypred_RF_train = model_RF.predict(X_train_scaled)
print('r-square_RF_Train: ', round(model_RF.score(X_train_scaled, y_train_array), 2))

mse_RF_train = metrics.mean_squared_error(y_train_array, ypred_RF_train)  
print('RMSE_RF_train: ', round(np.sqrt(mse_RF_train),4))

mean_absolute_error_RF_train = metrics.mean_absolute_error(y_train_array, ypred_RF_train)
print('MAE_RF_train: ', round(mean_absolute_error_RF_train, 4)) 

# Test metrics for Random Forest model
print('\nTest metrics for Random Forest model scaled data')
ypred_RF_test = model_RF.predict(X_test_scaled)
print('r-square_RF_test: ', round(model_RF.score(X_test_scaled, y_test_array), 2))

mse_RF_test = metrics.mean_squared_error(y_test_array, ypred_RF_test) 
print('RMSE_RF_test: ', round(np.sqrt(mse_RF_test), 4))

mean_absolute_error_RF_test = metrics.mean_absolute_error(y_test_array, ypred_RF_test)
print('MAE_RF_test: ', round(mean_absolute_error_RF_test, 4))

Training metrics for Random forest model using scaled data
r-square_RF_Train:  0.97
RMSE_RF_train:  6.9747
MAE_RF_train:  3.4844

Test metrics for Random Forest model scaled data
r-square_RF_test:  0.77
RMSE_RF_test:  22.2146
MAE_RF_test:  10.2689

feature_imp_RF = model_RF.feature_importances_

#relative feature importance  
rel_feature_imp = 100 * (feature_imp_RF / max(feature_imp_RF)) 
rel_feature_imp = pd.DataFrame({'features':list(X_train.columns),
                                'rel_importance':rel_feature_imp })

rel_feature_imp = rel_feature_imp.sort_values('rel_importance', ascending=False)


#plotting the top twenty important features
top20_features = rel_feature_imp.head(20) 

plt.figure(figsize=[20,10])
plt.yticks(fontsize=15)
ax = sns.barplot(x="rel_importance", y="features",
                 data=top20_features,
                 palette="Accent_r")

plt.xlabel("Relative Importance", fontsize=25)
plt.ylabel("Features", fontsize=25)
plt.show()

rel_feature_imp.head()

# GradientBoosting with non scaled data
# this model uses the default parameters of GB algorithm, while the estimators are changed to obtain the best fit 
model_GB_nonscale = GradientBoostingRegressor(n_estimators=500, random_state=60)

# Train the model
model_GB_nonscale.fit(X_train, y_train_array)

# Training metrics for Gradient Boosting Regressor model
print('Training metrics for Gradient Boosting Regressor model using scaled data')

ypred_GB_train = model_GB_nonscale.predict(X_train)
print('r-square_GB_Train: ', round(model_GB_nonscale.score(X_train, y_train_array), 2))

mse_RF_train = metrics.mean_squared_error(y_train_array, ypred_GB_train)
print('RMSE_GB_Train: ', round(np.sqrt(mse_RF_train), 4))

mean_absolute_error_RF_train = metrics.mean_absolute_error(y_train_array, ypred_GB_train)
print('MAE_GB_Train: ', round(mean_absolute_error_RF_train, 4))

#Test metrics for Gradient Boosting Regressor model
print('\nTest metrics for Gradient Boosting Regressor model using scaled data')

ypred_GB_test = model_GB_nonscale.predict(X_test)
print('r-square_GB_Test: ', round(model_GB_nonscale.score(X_test, y_test_array),2))

mse_RF_Test = metrics.mean_squared_error(y_test_array, ypred_GB_test)  
print('RMSE_GB_Test: ', round(np.sqrt(mse_RF_Test),4))

mean_absolute_error_GB_Test = metrics.mean_absolute_error(y_test_array, ypred_GB_test)
print('MAE_GB_Test: ', round(mean_absolute_error_GB_Test, 4))

Training metrics for Gradient Boosting Regressor model using scaled data
r-square_GB_Train:  0.99
RMSE_GB_Train:  3.1363
MAE_GB_Train:  2.3217

Test metrics for Gradient Boosting Regressor model using scaled data
r-square_GB_Test:  0.81
RMSE_GB_Test:  20.0925
MAE_GB_Test:  9.5355

#checking the feature importance for the Gradient Boosting regressor
feature_imp_GB = model_GB_nonscale.feature_importances_
rel_feature_imp_GB = 100 * feature_imp_GB / max(feature_imp_GB)
rel_feature_imp_GB = pd.DataFrame({'features':list(X_train.columns),
                                   'rel_importance':rel_feature_imp_GB})
rel_feature_imp_GB = rel_feature_imp_GB.sort_values('rel_importance', ascending=False)
rel_feature_imp_GB.head()

# Plot  feature importance for the Gradient Boosting regressor
top20_features_GB = rel_feature_imp_GB.head(20) 

plt.figure(figsize=[20,10])
plt.yticks(fontsize=15)
ax = sns.barplot(x="rel_importance", y="features", data = top20_features_GB, palette="Accent_r")
plt.xlabel("Relative Importance", fontsize=25)
plt.ylabel("Features", fontsize=25)
plt.show()

# Validating with a 10 fold cross validation for the Gradient Boosting models
y_array = y.values.flatten()

modelGB_cross_val = GradientBoostingRegressor(n_estimators=500, random_state=60) 

modelGB_cross_val_scores = cross_val_score(modelGB_cross_val,
                                           X, 
                                           y_array,
                                           cv=10,
                                           scoring='neg_mean_absolute_error')

print("All Model Scores: ", modelGB_cross_val_scores)

print("Negative Mean Absolute Error: {}".format(np.mean(modelGB_cross_val_scores)))

All Model Scores:  [ -6.8058649  -11.62520912 -11.02025161 -21.53356819  -4.63056795
 -41.11872138 -14.46560863  -9.15669676  -6.10330726  -4.9869393 ]
Negative Mean Absolute Error: -13.1446735095419

# Validating with a 10 fold cross validation for the Random forest models
y_array = y.values.flatten()

modelRF_cross_val = RandomForestRegressor(n_estimators=500, random_state=43)

modelRF_cross_val_scores = cross_val_score(modelRF_cross_val,
                                           X, 
                                           y_array,
                                           cv=10,
                                           scoring='neg_mean_absolute_error')

print("All Model Scores: ", modelRF_cross_val_scores)

print("Negative Mean Absolute Error: {}".format(np.mean(modelRF_cross_val_scores)))

All Model Scores:  [-10.648       -9.98169585 -10.22411982 -22.20942857  -4.53489401
 -37.69979724 -16.13404608 -12.30612963  -6.35166667  -4.42249074]
Negative Mean Absolute Error: -13.45122686038573

# Plotting a kernel density map of the predicted vs. observed data
plt.figure(figsize=[15,5])

# plotting the prediction
sns.kdeplot(ypred_RF_test, label = 'Predictions', color = 'orange')
y_observed = np.array(y_test).reshape((-1, ))
sns.kdeplot(y_observed, label = 'Observation', color = 'green')

# label the plot
plt.xlabel('No. of Airbnb listings per census tract', fontsize=15)
plt.ylabel('Density', fontsize=15)
plt.title('Density Plot: Predicted vs Observed', fontsize=15)
plt.xticks(range(0,500,25), fontsize=10)
plt.yticks(fontsize=10)
plt.legend(fontsize=15)
plt.show()

# Converting the predicted and observed values to dataframe and plotting the observed vs predicted
y_test_df = y_test.copy()
y_test_df['Predicted'] = (ypred_RF_test)  
y_test_df.head()

# plotting the actual observed vs predicted airbnb properties by tract
plt.figure(figsize = [25,12])
sns.set(style = 'whitegrid')
sns.lineplot(data = y_test_df, markers=True) 

#label the plot
plt.xlabel('Tract ID', fontsize=15)
plt.ylabel('Total No. of Airbnb', fontsize=15)
plt.title('Actual No. of Airbnb by Tract: Predicted vs Observed', fontsize=15)
plt.xticks(range(0,2000,100), fontsize=15)
plt.yticks(fontsize=15)
plt.legend(fontsize='x-large', title_fontsize='10')
plt.legend(fontsize=15)
plt.show()