Predicting real estate prices from the urban environment

Slide 1

Slide 1 text

The economic value of neighborhoods: Predicting real estate prices from the urban environment MARCO DE NADAI and BRUNO LEPRI

Slide 2

Slide 2 text

2 Let’s find a new house

Slide 3

Slide 3 text

3 Let’s find a new house

Slide 4

Slide 4 text

4 Let’s find a new house

Slide 5

Slide 5 text

5 Let’s find a new house

Slide 6

Slide 6 text

6 Let’s find a new house

Slide 7

Slide 7 text

7 Let’s find a new house

Slide 8

Slide 8 text

8 Let’s find a new house

Slide 9

Slide 9 text

9 Let’s find a new house

Slide 10

Slide 10 text

10 Let’s find a new house

Slide 11

Slide 11 text

11 Let’s find a new house

Slide 12

Slide 12 text

12 Let’s find a new house

Slide 13

Slide 13 text

How is the housing price composed?

Slide 14

Slide 14 text

The data 14 One year of listed properties in Immobiliare.it • 8 biggest Italian cities • 70,000 houses

Slide 15

Slide 15 text

The property: textual features Textual features about the listed property • Number of rooms • Square meters • Energy compliance • Garden (yes/no) • etc 15

Slide 16

Slide 16 text

The property is not an island 16

Slide 17

Slide 17 text

The neighborhood The fundamental geographical unit for: • individuals’ activities • social interactions Studied in urban planning, social science, criminology Usually defined through non-overlapping administrative boundaries (census) 17

Slide 18

Slide 18 text

The neighborhood/egohood 18

Slide 19

Slide 19 text

The neighborhood/egohood 19 So what about the features?

Slide 20

Slide 20 text

Neighborhood: walkability Walkable and pleasant neighborhoods have a direct impact on housing prices Online markets (e.g. Zillow) now show near restaurants and schools 20 Cortright, Joe. "Walking the walk: How walkability raises home values in US cities." (2009).

Slide 21

Slide 21 text

Neighborhood: walkability Walking distance to amenities (~ to Walkscore™) • Restaurants and bars (10) • Shops (5) • Schools • Grocery stores • Entertainment • Parks (3) * (n) = depth of choices 21

Slide 22

Slide 22 text

Neighborhood: walkability Walking distance to amenities (~ to Walkscore™) 22

Slide 23

Slide 23 text

Neighborhood: walkability Walking distance to amenities (~ to Walkscore™) We have a score of walkability for each house 23 0 500 1000 1500 2000 Distance 0.0 0.2 0.4 0.6 0.8 1.0 Walkability score

Slide 24

Slide 24 text

Neighborhood: urban theory Jane Jacobs has been one of the most influential people in urban planning Diversity => Urban vitality => Higher Housing price? 24 Jacobs, Jane. The death and life of great American cities. Vintage, 1961

Slide 25

Slide 25 text

Neighborhood: urban theory Four essential conditions: 1. Land use mix 25 Jacobs, Jane. The death and life of great American cities. Vintage, 1961 Land use mix

Slide 26

Slide 26 text

Neighborhood: urban theory Four essential conditions: 1. Land use mix 2. Small blocks 26 Jacobs, Jane. The death and life of great American cities. Vintage, 1961 Small blocks

Slide 27

Slide 27 text

Neighborhood: urban theory Four essential conditions: 1. Land use mix 2. Small blocks 3. Aged buildings 27 Jacobs, Jane. The death and life of great American cities. Vintage, 1961 Aged buildings

Slide 28

Slide 28 text

Neighborhood: urban theory Four essential conditions: 1. Land use mix 2. Small blocks 3. Aged buildings 4. Density 28 Jacobs, Jane. The death and life of great American cities. Vintage, 1961 Density

Slide 29

Slide 29 text

Neighborhood: urban theory Four essential conditions: 1. Land use mix 2. Small blocks 3. Aged buildings 4. Density 29 De Nadai, Marco, et al. "The death and life of great Italian cities: a mobile phone data perspective." WWW, 2016. OpenStreetMap + census!

Slide 30

Slide 30 text

Neighborhood: cultural capital Cultural capital influences housing prices and people behavior • Publishing • Video, radio, photography • Architecture • … And heavy industries (negative!) 30 Hristova, Desislava, Luca M. Aiello, and Daniele Quercia. "The New Urban Success: How Culture Pays." Frontiers in Physics 6 (2018): 27.

Slide 31

Slide 31 text

Neighborhood: security perception 31 Which place looks more secure?

Slide 32

Slide 32 text

Neighborhood: security perception 32 Salesses, Philip, Katja Schechtner, and César A. Hidalgo. "The collaborative image of the city: mapping the inequality of urban perception." PloS one 8.7 (2013)

Slide 33

Slide 33 text

• Learning safety perception, learned in Milan, Rome • Predict on our cities Neighborhood: security perception 33 De Nadai, Marco, et al. "Are safer looking neighborhoods more lively?: A multimodal investigation into urban life." ACM MM 2016.

Slide 34

Slide 34 text

Neighborhood and place: together! 34 PLACE Characteristics of the census cell NEIGHBORHOOD Objective and subjective characteristics HOUSE Characteristics of the house

Slide 35

Slide 35 text

The model

Slide 36

Slide 36 text

XGBoost 36 = ( + ) Neighborhood features (e.g. walkability) Property features (e.g. square feets) Housing price (ground truth) K-fold Cross-validation (with care!) Contiguity matrix

Slide 37

Slide 37 text

37 Results (errors) Model MAE MdAPE Property 148, 109 23,76%

Slide 38

Slide 38 text

38 Results (errors) Model MAE MdAPE Property 148, 109 23,76% Property + Neighborhood 104,586 15,44% THE NEIGHBORHOOD SHAPES PROPERTY PRICE BY ~60%!

Slide 39

Slide 39 text

39 Global features importance

Slide 40

Slide 40 text

40 How can we explain predictions? From a prediction we can go back and compute how each feature contributed to the prediction J: set of trees K: set of features

Slide 41

Slide 41 text

41 Can we explain predictions? House in Corso Grosseto, Turin

Slide 42

Slide 42 text

42 Can we explain predictions? House in Corso Grosseto, Turin THE MODEL IS INTERPRETABLE

Slide 43

Slide 43 text

43 Reproducibility Some of the principal competitors: • Make use of timeless, private, data • Do not share data • Do not share code

Slide 44

Slide 44 text

44 Reproducibility: Open model + code We created an “Open” model: • Make use only of Open data • Share the data • Share the code! https://github.com/denadai2/real-estate- neighborhood-prediction

Slide 45

Slide 45 text

45 Results (errors) Model MAE MdAPE Property 148, 109 23,76% Property + Neighborhood 104, 586 15,44% Property + Neighborhood (Open) 138, 929 18.02% GREAT!

Slide 46

Slide 46 text

No content

Slide 47

Slide 47 text

Why is it important 47 • We can better understand the market • Planning: no oasis in the desert • The surroundings and vitality is important • Responsive predictions without historical data • Gentrification?