Overview

1 Theme

The theme we would like to explore is Spatial Point Patterns Analysis.

2 Abstract

Spatial point patterns analysis (SPPA) studies the distribution of the points, whether the distribution is random or clustered. This form of analysis can be very useful in the evaluation of events and we would be able to investigate whether there are any dependency relationships between different point distributions to make a comparison and conclusion.

Another kind of spatial point patterns analysis is called network constrained spatial point patterns analysis (NetSPPA) which allows us to analyse if the distribution of the spatial point events are affected by a network or whether the spatial point events occur alongside a network. 

The application our team developed allows tourists, policy makers and community members to visualise and analyse the distribution of Airbnb listings in Singapore on a map. The app can be used to help better understand the impact of Airbnbs on the local housing market, improve urban planning and enhance tourism development. With our app, users can select and modify their inputs to conduct both SPPA and NetSPPA. 

For spatial point patterns analysis, we would like to find out if the Airbnb locations in Singapore are randomly distributed throughout the country and if not, where are the locations with higher concentrations of Airbnb. Also, at these locations of higher concentration, do the Airbnb locations co-exist with other point events like train stations, hotels, etc.? Users can use Kernel Density Estimation, G-Function, F-Function, Cross K-Function and Cross L-Function.

For network constrained spatial point patterns analysis, we would like to discover whether the distribution of the Airbnb locations are affected by the road network in Singapore. Through these analyses, we can investigate whether the distribution of Airbnb locations in Singapore are affected by point events or the road network. Users can use Network Kernel Density Estimation, K-Function and Network Cross K-Function.

3 Problem & Motivation

Our team decided to develop an application that allows users to analyse Airbnb listings in Singapore and their relationships with other points of interest (MRTs, bus stops etc.) as there is a lack of such apps for Singapore Airbnb listings.

After the COVID pandemic, with travelling becoming the norm and Singapore as one of the most popular cities to travel to in the world, Singapore Airbnb listing information would deem to be extremely valuable. Moreover, with the high cost of living in Singapore, many tourists would turn to cheaper accommodation options like Airbnb rentals.

In order to provide a good accommodation experience for such tourists, it is crucial to analyse the current Airbnb listings in Singapore to study the regions where Airbnb listings are most prevalent and why.

4 Project Objectives

In this project, we would like our Shiny web application to help users:

1. Visualise distributions and realise the benefits of spatial point patterns analysis
2. Conduct spatial point patterns analysis (Kernel density estimation, G-Function, F-Function, Cross K-Function, Cross L-Function)
3. Conduct network constrained spatial point patterns analysis (NetKDE and K-Function, Cross K-Function)
4. Use all the insights gathered from the analysis and models to make practical decisions

5 Main Features

The main features for the analytical tools are:

• Spatial Point Patterns Analysis
  • Filters
    • Different variables for analysis
    • Various bandwidth selection methods (diggle, CvL, scott, ppl)
    • Various Kernel Methods (gaussian, epanechnikov, quartic, disc)
    • Various Statistical Functions (G/F/Cross K/Cross L-Functions)
  • Sliders
    • For fixed bandwidth selection method
  • Input
    • Number of simulations
• Network Constrained Spatial Point Patterns Analysis
  • Filters
    • Different variables for analysis
    • Various kernel methods (gaussian, epanechnikov, quartic, disc)
    • Various NKDE methods (simple, discontinuous, continuous)
  • Input
    • Number of simulations

6 Data Sources

Spatial Point Patterns Analysis

• URA 2019 Master Plan Planning Subzone Boundary Data (.shp) - By Prof Kam in In Class Ex 9
• Airbnb Locations in Singapore (.csv) - We will get the point events via the longitude and latitude
• MRT Stations in Singapore (.shp)
• Bus Stops in Singapore (.shp)
• Tourist Attractions in Singapore (.shp)
• Shopping Malls in Singapore (.csv) - We will use this list of shopping mall names to get the point events using OneMapSG API
• 7-11 Stores in Singapore - We will extract the list of stores (name, address, postal) into xlsx, then locate their point events using OneMapSG API
• Hotel Locations in Singapore (.kml)
• Universities and Colleges in Singapore - We will extract the list of universities and colleges (name) into xlsx, then locate their point events using OneMapSG API

Network-Constrained Spatial Point Patterns Analysis

• Road Network of Singapore (Road Section Line) (.shp)
• Airbnb locations in Singapore (.csv) - We will get the point events via the longitude and latitude

7 Approach / Methodology

• Data Preparation
  • Gather data from various sources
  • Import geospatial and aspatial data
  • Data wrangling
• Exploratory Data Analysis
  • Visualisation of the data
  • Checking for outliers, anomalies, missing values, noise
  • Further data wrangling (if necessary)
• Spatial Point Patterns Analysis
  • First-order analysis
    • Kernel Density Estimation
  • Second-order analysis
    • G-Function
    • F-Function
    • Cross K-Function
    • Cross L-Function
• Network Constrained Spatial Point Patterns Analysis
  • First-order analysis
    • NetKDE Analysis
  • Second-order analysis
    • K-Function
    • Cross K-Function

8 Literature Review

8.1 Related Works

Seniors’ Works - Spatial Pointers

Statistical Functions Seniors Employed

When we explored the theme of Spatial Point Patterns Analysis, we were inspired by one of the seniors’ Shiny application that allows users to upload their data for SPPA and NetSPPA. For their application, they did SPPA mainly on McDonald’s outlets in Singapore and NetSPPA mainly on childcare centres in Punggol, Singapore. We referenced the SPPA section where 4 main types of statistical analysis were used:

• Kernel Density Estimation: Visualise which area(s) have the highest intensity of points density
• G-Function: Determine if the point events resemble clustered, dispersed or random distribution using cumulative frequency distribution of the nearest neighbour distance of a point pattern
• F-Function: Determine if the point events resemble clustered, dispersed or random distribution by calculating the minimum distance from a random point to the original points
• Cross L-Function: Determine if there is correlation between two types of point events by calculating the number of points in a given distance

Statistical Functions Our Team Employed

For our application, we similarly implemented the following functions:

• Kernel Density Estimation: Visualise which location(s) in the selected zone have the highest intensity of Airbnb points density
• G-Function: Determine if the Airbnbs in the selected zone resemble clustered, dispersed or random distribution using cumulative frequency distribution of the nearest neighbour distance of a point pattern
• F-Function: Determine if the Airbnbs in the selected zone resemble clustered, dispersed or random distribution by calculating the minimum distance from a random point to the original points
• Cross K-Function: Determine if there is correlation between two types of point events (eg. Airbnbs & MRTs, Airbnbs & Bus stops etc.) by calculating the number of points in a given distance
• Cross L-Function: Determine if there is correlation between two types of point events (eg. Airbnbs & MRTs, Airbnbs & Bus stops etc.) by calculating the number of points in a given distance (Standardised version of Cross K-Function)

Seniors’ Function Implementations VS Our Teams Function Implementations

For the seniors’ project, they did not include Cross K-Function which was understandable since Cross K-Function and Cross L-Function are extremely similar - given that the latter was derived from the former through standardisation.

However, for our project, we wanted to include more functions so that technically inclined individuals who are keen to explore the minute differences between the 2 mentioned functions are free to do so.

Seniors’ Works - Signal

Statistical Network Functions Seniors Employed

When developing our application, we were inspired by one of the seniors’ Shiny applications which explored the use of Network-constrained Spatio-temporal Analysis for Traffic Accidents in the UK. In that application, 4 main types of statistical analysis were used:

• Network-Constrained Kernel Density Estimation: Visualise which segment(s) of the road have the highest intensity of traffic accident points or casualty points along the network
• Network-Constrained K-Function: Determine if there is spatial correlation between accident points on a linear network using geometrically corrected K-Function
• Network-Constrained Cross K-Function: Determine if there is correlation between accident points and variables selected
• Network-Constrained Cross Pair Correlation Function: Determine if there is correlation between accident points and variables selected

Statistical Network Functions Our Team Employed

For our application, we similarly implemented the following Network-related functions:

• Network-Constrained Kernel Density Estimation: Visualise which segment(s) of Rochor’s street network have the highest intensity of Singapore Airbnb listing points or amenity points (MRT, bus stops etc.) along the network
• Network-Constrained K-Function: Determine if there is spatial correlation between Rochor Airbnb listing points on a linear network using geometrically corrected K-Function
• Network-Constrained Cross K-Function: Determine if there is correlation between Airbnb listing points in Rochor and amenity variables selected

We focused on Rochor as Rochor has a significant number of Airbnbs since it is the zone with the 4th most number of Airbnbs. Additionally, Rochor has enough point events for the other amenity points (MRT, bus stops, etc.) which will allow us to draw statistical conclusions with the Airbnbs in Rochor. There were other zones with higher numbers of Airbnbs but we did not choose them as they only had 1 point event for the other amenity points which is not enough to draw statistical conclusions especially for Network-Constrained Cross K-Function.

Seniors’ Network Function Implementations VS Our Team’s Network Function Implementations

For the seniors’ project, since they were analysing accidents, they focused on analysing factors that could affect accident rates in specific regions of the UK. In some of their network features, they had filters to control the time, environment (weather) and casualty.

For our project, since we were analysing Singapore Airbnb listing points, we focused more on studying the different types of amenity points around. This is because for our project, isolating and studying factors like time and weather would be less appropriate as Airbnb listings tend to be much more influenced by factors like amenities instead.

For the seniors’ implementation, they implemented less technical adjustments like time, environment (weather) and casualty. Since we wanted our application to be well-used by technically inclined experts, we introduced more technical adjustments. For instance, for the Network-Constrained Kernel Density Estimation under our app’s NetSPPA tab, besides allowing users to select different variables (Airbnbs, MRTs, bus stops), we allowed users to control more technical features such as ‘Kernel Smoothing Input’ and ‘Method’. At the bottom of the Network graphs generated, we’ve also additionally included descriptions to aid the users in interpreting the graphs. This helps less technical users to comprehend the results more easily.

8.2 Other Studies

Spatial Distribution of Airbnb Supply in Los Angeles

• Summary
  • This study investigates the spatial patterns of Airbnb supply. In the city of Los Angeles during the years 2014 to 2019. This study integrates exploratory spatial data analysis (ESDA) and explores the determinants of spatial patterns of Airbnb supply in 2019 with spatial regression models. The results of the spatial autocorrelation analysis show that the spatial distribution of Airbnb supply in Los Angeles is positively clustered and shows a geographical logistical trend over time.
  • This paper used socio-economic data including employment, population, income, education, and housing units.
• Learning Points
  • ESDA (to describe spatial distribution + quantitative models)
    • For quantitative models under ESDA, the paper covered Global Moran’s I statistic and the Anselin Local Moran’s I (LISA) statistic to measure spatial autocorrelation.
  • Distance between points eg. the city center and Hollywood was calculated using GIS tools. The POI (point of interest) is a specific point location that someone may find useful or interesting, which to some extent can be used to determine the attractiveness of a place. The POI data were obtained from GEOFABRIK, where OpenStreetMap data was collected, showing a total of 5,424 POI in Los Angeles
• Linking Back To The Project
  • We could use the ESDA methods in the paper and explore the use of POI. The paper showed that Airbnb supply is positively associated with the number of POI.

Point Pattern Analysis on the 1854 cholera outbreak in London

• Summary
  • This study talks about the methods Point Pattern Analysis (PPA) and focuses on a case study which is the 1854 cholera outbreak in London
• Learning Points
  • PPA plays an important role in conducting exploratory analysis on point data
  • Commonly used PPA methods are 
    • descriptive statistics, 
    • distance-based measures (nearest-neighbour distance, distance functions) 
    • and density-based measures (Quadrat density, Kernel density).
• Linking Back to the Project
  • Like the study, we can implement distance-based measures like the distance functions (G/F/K Function) and also density-based measures (Kernel density) for our Spatial Point Patterns Analysis 

Network constrained and classified spatial pattern analysis of healthcare facilities and their relationship with the road structure: a case study of Thiruvananthapuram city

• Summary
  • Network-based location analysis was carried out to find the distribution pattern of the healthcare facilities in Thiruvananthapuram city and also the distribution of healthcare facilities with reference to the network distance (to know whether the distribution pattern of healthcare facilities is clustered or random, whether they tend to follow a specific direction or whether they are clustered around any specific point)
• Learning Points
  • Centrographic analysis (using Mean Center, Standard Deviational Ellipse and Average Nearest Neighbour) was performed to study the distribution pattern of the healthcare facilities, whether the healthcare facilities locations have a relationship with the road network
  • To find the distribution of medical stores with respect to the hospitals, cross K function was performed
  • Kernel Density Estimation technique was applied to modify the centrality values and the vector points to a basic raster framework.
  • ANN uses the distance between each point entity and its closest neighbouring entity to predict if the point pattern is random, clustered or dispersed.
• Linking Back to the Project
  • We can implement cross K function to investigate the location of Airbnbs are influenced by the road network
  • We could use the ANN ratio (average nearest neighbour ratio) to reveal if a point pattern is a clustered distribution. If the ratio >1, we can conclude that the pattern is more dispersed than random. A value of 1 indicates a random distribution. To check the statistical significance of the ANN ratio, we will calculate the z-score in order to conclude confidently if the pattern is indeed random, clustered or dispersion.
    • 

9 Storyboard

Home Page

Visualisation Page

Tmap Page

Spatial Point Patterns Analysis Page

Network Constrained Spatial Point Patterns Analysis Page

10 Application Architecture

11 Timeline