XccessPoint Proposal

Proposal

Poster

Our Application

Research Paper

The Team

Project Motivation

In recent years, increasing attention has been paid to the issue of inequality in Singapore among parliament discussions and social policy studies. “This is what inequality looks like.” You Yenn Teo’s recent best seller book in 2018 uncovers the heightened tension on social inequalities in Singapore through illuminating an ethnographic presentation of the experiences of the less privileged Singaporeans. Moreover, in the recent Commitment to Reducing Inequality Index 2018 conducted by International Confederation Oxfam International, Singapore was ranked as one of the bottom 10 countries worldwide in inequality reduction.

The current state of inequality has motivated to use to delve deeper into the spatial inequality in Singapore which has not been widely examined in the past. We hope to understand inequality by examining the accessibility to many key essential facilities for an ordinary Singaporean living in Housing Development Board (HDB) units. The improvement in visibility to geospatial inequality through our application could provide policy makers a more justified and structured approach for strategizing future plans in mitigating inequality in different neighbourhoods.

Project Description

Our project would like to develop geographical accessibility and spatial interaction model to study the accessibility of HDB units to facilities in Singapore. The aspects of accessibility to look into includes the distance to healthcare facilities (General Practitioner Clinics, Polyclinics and Hospitals), transportation infrastructure (MRT stations and Bus Stops), Schools, Police Stations, and Hawker Centres for all HDBs in different regions, planning areas and planning subzones. We hope to develop an accessibility study tool through Analytic Hierarchy Process and enable urban planners leverage on this open-source, interactive, reproducible and highly accessible tool to better strategize urban planning policies.

Project Objectives

In summary, through our project, we aim to:

1. Study the spatial distribution of facilities and HDB units
2. Identify the areas with poor accessibility to essential facilities
3. Analyze and highlight regions with higher needs for certain facilities
4. Indicate whether there is a substantial difference in accessibility to facilities across regions, planning areas and planning subzones
5. Provide customizable input parameters on pairwise comparisons of facilities, thus allowing policy makers to generate overall Analytic Hierarchy Process accessibility score based on their prioritie
6. Evaluate results of analysis and provide recommendations for urban planners to enhance the accessibility to different facilities

Data Sources

Dataset	Description	Data Type	Source(s)
Singapore Regions	To facilitate urban planning, the Urban Redevelopment Authority (URA) divides Singapore into 5 regions, namely Central, West, North, North-East and East Regions.	SHP	Data Source
Singapore Planning Area	Indicative polygon of planning area boundary. To facilitate urban planning, the Urban Redevelopment Authority (URA) divides Singapore into 55 planning areas	SHP	Data Source
Singapore Planning Subzone	Indicative polygon of subzone boundary. The Planning Regions are divided into smaller Planning Areas. Each Planning Area is further divided into smaller subzones which are usually centred around a focal point such as neighbourhood centre or activity node.	SHP	Data Source
HDB	List of HDB location via postal code	CSV	Data Source
School facilities	List of education facilities in Singapore	CSV,KML	Data Source Data Source
Government Markets Hawker Centres	Contains Address of Hawker Centres in Singapore	KML	Data Source
Heathcare Facilities	Contains Address to Healthcare Facilities in Singapore	Website Information	Data Source Data Source
LTA Mrt station	The layer contains the locations of MRT station exits.	KML	Data Source
Bus Stops	All bus stops, bus interchanges, bus terminals in Singapore.	SHP	Data Source

Literature Review

Literature review of relevant research paper on spatial analysis of accessibilities are conducted to enhance our project methodology.

1.Site Suitability Evaluation for Ecotourism Using GIS & AHP: A Case Study of Surat Thani Province, Thailand

Study Objective:

This paper aims to identify and prioritize the potential ecotourism sites using Geographic Information System (GIS) and Analytical Hierarchy Process( AHP) in Surat Thani Province. The factors in consideration for suitability for the land ecosystems include landscape/naturalness, wildlife, topography, accessibility and community characteristics.

Visualization:

Figure 1: Suitability Map for Ecotourism in Surat Thani Province in Thailand

Figure 2:Schematic Representation of the Methodology

Figure 3:AHP Matrix for Pairwise Comparisons and the Consistency Ratio Estimation

Methodology

1. Determination of Weights using AHP

AHP is one extensively used Multi-Criteria Decision Making technique (developed by Saaty in 1980) used in structural decision making process for complex problems that involves multiple criteria across different hierarchical levels. Pairwise comparisons method is used to compare the criteria and allow for evaluation of relative significance of all parameters. Expert opinions were taken into consideration for the comparisons. Pairwise comparison uses a scale of 1 to 9 which 1 means having equal importance while 0 means having extreme importance. Reciprocal pairwise comparisons is used for opposite comparison of facilities.

2. Factoring in Decision Making Inconsistency

To ensure consistent judgement of decision makers, AHP efficiency criteria are measured by Consistency Relationship (CR = Consistency Index/Random Index). If CR is smaller than 0.10, degree of consistency will be fairly acceptable. Otherwise if it exceeds 0.10, inconsistencies exist in the evaluation process and we need to reject the pairwise comparisons and reiterate the process.

3. Land Suitability Assessment

The total suitability score “Si” for each land unit (i.e. each raster cell in the map) was calculated from the linear combination of suitability score obtained for each factor and criteria involved.

where “n” is the number of factors, “Wi” is the multiplication of all associated weights in the hierarchy of “ith” factor ( as seen in Table 5) and “Ri” is a rating given for the defined class of the “ith” factor found on the assessed land unit

Learning Point:

AHP will be an highly effective methodology for us to reduce the complexity in computing overall accessibility score by structurally factoring the pairwise comparisons of all facilities. Consistency Ratios need to be factor in too.

Linear weighted combination of accessibility score could be adopted for our study

Caveat:

As this analysis is done on a proprietary software (ArcGIS 9.3), it is difficult for researchers to replicate the methodology of the research paper unless they have access to such software. As we aim to provide urban planners an open-source and easily reproducible application through R programming, there is a need to find similar packages for such methodology on R programming.

2.Estimating Spatial Accessibility to Facilities on the Regional Scale: an Extended Community-based Interaction Potential Model

Study Objective:

The study aims to leverage on measurements of spatial accessibility to regional facilities using aggregated data.The set of facilities includes three types of food outlets on the regional level at Bas-Rhin department,, namely hyper/supermarkets, grocery stores and bakeries.

Visualization:

Figure 4:Maps of number of grocery stores (left) and potential accessibility surface(right)

Methodology

1. Frequency Count of Opportunities within a Given Neighborhood

Frequently used indices based on count was first illustrated to give users a quick overview of the spatial distribution of facilities. This is known as container index which overcomes the limitation of individuals choosing only the nearest facility for consumption and actually consider all available opportunities within a neighborhood.

2. Community-based Interaction Potential Model

This model take into account of difference in urbanization level in the region when computing the accessibility level. As accessibility is a multi-dimensional concept, travel behaviours of the population are factored in. Kernel density estimation and Enhanced Two-Step Floating Catchment Area Method are used for accessibility assessment. These methods consider demand (population) and supply (health practitioners) side as well as travel impedance specification by assigning higher weight to opportunities in nearer region.

Learning Point:

Importance of Aggregating Data without consideration of administrative boundaries

To avoid administrative boundaries from limiting the number of closest facilities for each HDB units, administrative boundaries should be ignored when aggregating data. This is more realistic and precise estimation of accessibility levels such that we will not have areas with null accessibility.

Data Availability

To effectively implement such customized potential model for spatial accessibility analysis, we need demand, supply of facilities and household characteristics at each HDB units including travel impedance.

Caveat:

The study uses population of the region as the base reference to calculate the accessibility to the facilities. However, it failed to take into account of comparison with specific entities (such as HDB flats in our study).

2. The models developed in this study were implemented in the XLISP-STAT programming environment and ArcGIS 9.2 was used for the mapping visualizations. To adopt similar methodology, we need to ensure our programming language have similar functions.

3.Enhancing Healthcare Accessibility Measurement using GIS: A Case Study in Seoul, Korea

Study Objective:

This paper proposes a new method, Seoul Enhanced 2-Step Floating Catchment Area (SESSFCA) to study the accessibility of citizens to healthcare facilities in Seoul. Maintaining accurate and up-to-date information on healthcare accessibility allows the relevant government bodies to strategize future improvements and this includes expansion of healthcare infrastructure, effective labor distribution, alternative healthcare options for the regions with low accessibility, and redesigning the public transportation routes and schedules.

Visualization:

Figure 5:Comparison of Three Methods for Healthcare Accessibility Measurements

Figure 6:Accessibility to Private Healthcare

Figure 7: Box Plot Showing Average Accessibility to Private Healthcare by Districts

Methodology

1. Seoul Enhanced 2-Step Floating Catchment Area (SE2SFCA)

Where S1 and S2 are the standard number of physicians for distinguishing healthcare facilities between a regular hospital, hospital complex and large hospital complex.

This methodology is customized to Seoul city as it factors the fact that the population density is higher and the average distance between healthcare-service locations tends to be shorter than the typical North American or European cities. In addition, Seoul has a higher hospital density than other typical cities. In Korea, a healthcare facility is categorized into regular hospital, hospital complex and large hospital complex in accordance with the size and the number of provided medical specialties. The customized method proposed is more effective and realistic in identifying the regions with weaker accessibility.

2. Critical Distance Boundary Determination

Critical distance boundary (Dt) calculated from the critical travel time, is modeled as a function considering the travel mode of each population. ci is the number of private vehicles per person at population location i, and vc and vp are average speeds of private vehicle and public transportation modes, respectively. In the case of using the public transportation, the travel time boundary is also penalized by subtracting the waiting time (tw) from t. The critical distance boundary sets a distance buffer by factoring in the different socioeconomic status of individuals such as their vehicle possession.

3. Accessibility Calculation

Accessibility to healthcare is determined by geographical distances to service organizations, travel time, available modes of transport, population by region, average car ownership per person, average waiting time for public transportation and Income Differential Indices. The measurement is also separated to accessibility measurement for private and public healthcare facilities.

Learning Point:

To effectively implement such customized potential model for spatial accessibility analysis, we need demand, supply of facilities and household characteristics at each HDB units including vehicle ownership, population breakdown, income level and etc.

Boxplot can be utilized to show attribute distribution of accessibility score.

Caveat:

As this analysis is done on a proprietary software(QGIS), it is difficult for researchers to replicate the methodology of the research paper unless they have access to such software. There is a need to source for relevant features/packages in combining location attributes and calculation of important metrics.

Team's Approach

The table below shows a overall summary of our methodology

1. Data Preparation
   • Perform geocoding (Google API)
   • Change coordinate system to SVY21 (EPSG:3414)
   • Merging boundaries on different levels (region, planning area or subzone)
2. Set Accessibility Radius
   • Since comparing across facilities all over Singapore for each house is inefficient and computationally expensive, we will create a buffer of ~1km around each house and calculate distance to facilities within this radius only.
   • If no facilities are found, accessibility radius will be increased to beyond 1km
   • Next, compute minimum distance for each house to each type of facility using Euclidean Distance
3. Accessibility Calculation using Analyic Hierarchy Process
   • User will need to enter his/her priorities for the facility types in a pairwise matrix from -9 to 9
   • Calculate consistency
   Consistency Ratio (CR) = Consistency Index(CI)/ Random Index(RI) CI = (λmax -n) / (n-1) Where λmax is the maximum eigenvalue of the pairwise comparison vector and n is the number of attributes. Random Index is the mean of the resulting consistency index based on the order of the matrix
   • If the criteria entered by user is not consistent(CR>0.1), user will be prompted to reenter the priority
   • If consistent, weight for each facility will be calculated
   • Calculate weighted sum where each weight is multiplied by minimum distance to that facility.
   AHPi = SUM(Wj x Dij) where i is each house and j is each facility. Wj is weight of facility j and Dij is minimum distance from house i to facility j
   • This weighted sum forms the AHP score which is then plotted in the graph. The lower the AHP score, the higher the accessibility

Application Prototype

Data Page

Our team first plans to display the locations of all the facilities in Singapore on a plot. We will also plot a Chrolopleth map where colour of the subzone represents the number of facilities in that subzone

Accessibility Page

Then we will calculate distance from each HDB to the facility selected (i.e. Healthcare Services in this case). We will plot it in a map of the region/planning area/subzone selected. We will also calculate boxplots comparing planning areas and subzones in the region selected.

Analytic Hierarchy Process

Finally, we will calculate AHP for all the houses based on the priority entered and plot it according to the boundaries selected.

Tools and Data Architecture

Technical Challenge

No.	Key Challenges	Description	Solution(s)
a.	Data Collection	Many of the datasets required are not widely published online. There is a need to find alternative sources or ways to derive the data indirectly.	Obtain data from other existing websites or alternative government websites. For instance, HDB units is hard to retrieve and hence we downloaded the information from Energy Market website eventually as almost all units have energy consumption.
b.	Varying Level and Type of Dataset Atrributes	Having considered HDB units and various facility types, it is difficult to combine all due to the different data formats. Some of the datasets only provided addresses or KML files without proper data descriptions. The datasets also have varying coordinate reference systems.	Perform Geocoding for datasets with addresses Scrap relevant data attributes from datasets with KML files Standardize all CRS to SVY21 (EPSG:3414) Ensure all datasets are at the lowest aggregated level (planning subzone levels)
c.	High Computational Complexity for Distance Calculation	Since we had to calculate distances between thousands of HDBs with thousands of other facilities, we could not successfully compute them all due to time constraint and limitation in our computer processing speed and power.	Simplified the approach by only considering Euclidean Distance as the dataset is large with multiple facilities and HDB units. Use a distance radius buffer of 1km around each HDB and only calculated the distance to the facilities in the 1km. Range is extended only if no facilities are found within the range.
d.	Slow loading of Application due to Extensive Number of Spatial Points	As we have six facility types together with 8000 plus HDB units, usage of functions (tmapicons) is not ideal while plotting of certain plots are very slow, cluttered and not insightful.	We reduced the scope such that we plotted icons only when a user selected a region/planning area/subzone so that reduced the number of icons to plot Simplified and automated parts of the code to reduce complexity to minimize time taken to plot various visualizations
e.	Difficulty in Applying Sophisticated Spatial Potential Models	As not all information on supply and demand of our facility types (e.g. Police Stations, Hawker Centres) can be found online or derived, it is difficult to apply such models that require substantial information.	Explore other accessibility models through literature review and eventually considered minimum distance approach AHP model is further considered to enhance the credibility of accessibility score by factoring in expert opinions.

Project Task Allocation

Task	Description	Developer	Expected Completion Date	Status
Interim Project Proposal	Finalise storyboard and problem statement Finish background survey/initial literature review Discuss technical challenges, milestone and task allocation	All	Week 9	Done
Data Collection	Finalize datasets to be used Confirm facilities to be considered Check for data quality (credibility, aggregation level, accuracy)	Raynie	Week 9	Done
Data Preprocessing	Geocoding Scraping of strings from KML files Conversion of CRS systems and merge into boundaries	Shubham	Week 10	Done
AHP Research	Research on AHP calculation and study Literature Review	Kaelyn	Week 10	Done
Exploratory Data Analysis Plots	Plot facility spatial distribution for all facilities and houses	Kaelyn	Week 10	Done
RShiny Dashboard Prototype	Propose RShiny Template Design initial user interface	Raynie	Week 11	Done
Distance Radius Measurement	Create a buffer of 1km around each house and identify close facilities	Shubham	Week 11	Done
Minimum Distance Calculation	Calculate Euclidean Distance for all facilities Calculate Network Distance for all facilities	Shubham	Week 11	Successful for Euclidean Unsuccessful for Network
Visualizations Plotting	Plot barcharts/boxplots and other comparative plots	Kaelyn	Week 11	Done
Analytic Hierarchy Process Calculation	Confirm the methodology for AHP Calculation Finalize Code and Package for AHP	Kaelyn	Week 11	Done
RShiny Dashboad Finalization	Add Filter Selection and Finalize Rshiny Widgets Build dashboard pages based on visualizations	Raynie	Week 12	Done
Poster	Poster content and design	Kaelyn	Week 12	Done
Accessibility Plot	Plot accessibility plot for AHP and all facilities	Shubham	Week 12	Done
Application Interface Design	Beautify maps and relevant visualizations	Kaelyn	Week 13	Done
Final R Shiny User Experience Improvement	Make the functions and features more user-friendly Improve sequence of filter and toggle based on feedbacks from presentation	Raynie	Week 14	Done
Final Application User Interface Design	Improve Aesthetics and add additional customisation	Kaelyn	Week 14	Done
Artifact Final Submission	Test and debug	All	Week 14	Done
Research Paper	Ensure all components covered	All	Week 14	Done
Final Update on Wiki Page	Ensure all links and information are updated	All	Week 14	Done

References

1. Leong, C. (2018, May 30). Stay connected. Retrieved from https://www.ipscommons.sg/inequality-has-a-geographic-dimension-between-and-within-neighbourhoods-in-singapore/
2. Chua, M. (Ed.). (2018, February 18). Inequality is a threat - name it, and face it. Retrieved from https://www.straitstimes.com/singapore/inequality-is-a-threat-name-it-and-face-it
3. Here's why Singapore is one of the worst countries in reducing inequality. (2018, October 10). Retrieved from https://sbr.com.sg/economy/news/heres-why-singapore-one-worst-countries-in-reducing-inequality
4. Y., Byon, Y., & Yeo, H. (2018). Enhancing healthcare accessibility measurements using GIS: A case study in Seoul, Korea. Retrieved from https://doaj.org/article/5ee6f964387d4f2786136d21147dcd5e
5. Bunruamkaew, K. & Murayama, Y. (2011, nd). Site Suitability Evaluation for Ecotourism Using GIS & AHP: A Case Study of Surat Thani Province, Thailand. Retrieved from https://www.sciencedirect.com/science/article/pii/S1877042811013474#!
6. Salze, P., Banos, A., Oppert, J., Charreire, H., Casey, R., Simon, C., . . . Weber, C. (2011, January 10). Estimating spatial accessibility to facilities on the regional scale: An extended commuting-based interaction potential model. Retrieved from https://ij-healthgeographics.biomedcentral.com/articles/10.1186/1476-072X-10-2

Comments

Feel free to comments, suggestions and feedback to help us improve our project!