About This Project

How we built an interactive intelligence map of Indian cuisine across Greater Melbourne, using open government data, place APIs, and census demographics.

1,437

Restaurants

361

SA2 Areas

Data Sources

40%

Peak Indian Pop.

Data Sources

ABS Census 2021 — Demographics

Government

The Australian Bureau of Statistics Census 2021 provides population demographics at the SA2 (Statistical Area Level 2) geographic level. We extract three key datasets from the General Community Profile DataPack for Victoria.

Coverage: 361 SA2 areas across Greater Melbourne

Format: CSV in ZIP DataPack (~13 MB)

Fields Used

G09G — Country of Birth: P_India_Tot — total persons born in India per SA2
G01 — Selected Person Characteristics: Tot_P_P — total population (both sexes) per SA2
G02 — Selected Medians and Averages: Median_tot_hhd_inc_weekly — median weekly household income, converted to annual (* 52)

SA2 Boundaries

SA2 polygon boundaries are fetched from the ABS ArcGIS REST API at geo.abs.gov.au, filtered to the Greater Melbourne GCCSA region. Geometries are simplified (Douglas-Peucker tolerance ~100m) for web rendering.

API: ABS ASGS 2021 SA2 MapServer

DataPack: ABS Census DataPacks

Google Places API — Restaurant Supply & Ratings

API

The Google Places API (New) provides the most comprehensive restaurant data, including ratings and review counts. We search for "Indian restaurant" across 30 key Melbourne suburbs, covering both inner city and outer growth corridors.

Coverage: 30 suburbs, ~414 unique restaurants

Max per suburb: 20 results (API limit)

Fields Used

displayName — restaurant name
location — latitude and longitude
rating — average star rating (1–5)
userRatingCount — total number of reviews
formattedAddress — street address
priceLevel — price bracket (0–4)

Suburbs Searched

Dandenong, Clayton, Truganina, Tarneit, Point Cook, Glen Waverley, Craigieburn, Werribee, Springvale, Footscray, Melbourne CBD, Epping, Berwick, St Albans, Caroline Springs, Hoppers Crossing, Deer Park, Sunshine, Narre Warren, Cranbourne, Pakenham, Box Hill, Preston, Reservoir, Melton, Richmond, South Yarra, Roxburgh Park, Wyndham Vale, Williams Landing.

API: Google Places Text Search (New)

Foursquare Open Source Places — Restaurant Supply

Open Data

Foursquare's Open Source Places dataset on HuggingFace contains ~100M+ places globally. We download only the Australian Parquet partitions (files 98 & 99, ~240 MB total), filter to the Melbourne bounding box, and identify Indian restaurants by name keywords.

Coverage: ~1,060 Indian restaurants in Greater Melbourne

Format: Parquet via HuggingFace Datasets API

Fields Used

name — place name
latitude / longitude — coordinates
locality — suburb name
postcode — postal code
address — street address
fsq_category_ids — Foursquare taxonomy category IDs

Filtering Method

Indian restaurants are identified by matching name keywords (indian, curry, dosa, tandoori, biryani, masala, sri lanka, nepal, etc.) and then filtered to exclude non-restaurant places (grocery stores, taxis, salons) using both category IDs and name exclusions.

Dataset: foursquare/fsq-os-places

License: Apache 2.0

City of Melbourne CLUE — CBD Restaurants

Government

The Census of Land Use and Employment (CLUE) dataset from Melbourne's Open Data portal provides official business registration data for cafes and restaurants within the City of Melbourne municipality (CBD and immediate surrounds only).

Coverage: ~36 Indian restaurants in CBD area

Limitation: City of Melbourne LGA only (not Greater Melbourne)

Fields Used

trading_name — business trading name
latitude / longitude — coordinates
business_address — street address
number_of_seats — seating capacity
clue_small_area — sub-area within the municipality
census_year — year of survey (latest: 2023)

API: Melbourne Open Data (OpenDataSoft)

Methodology

Fetch SA2 Boundaries

Download 361 SA2 polygon boundaries for Greater Melbourne from the ABS ArcGIS REST API. Simplify geometries for web display.

Join Census Demographics

Download the ABS Census 2021 DataPack for Victoria. Extract Indian-born population (G09G), total population (G01), and median income (G02). Join to SA2 boundaries by SA2 code. Compute india_pct = india_born / total_pop * 100.

Collect Restaurant Data

Fetch Indian restaurants from three sources: Foursquare (Parquet download + keyword filter), Google Places API (text search across 30 suburbs), and CLUE (Melbourne Open Data API). Deduplicate across sources by name + location proximity.

Cuisine Classification

Each restaurant is classified into a sub-cuisine by keyword matching on its name: South Indian, North Indian, Sri Lankan, Nepalese, Bangladeshi, or General Indian.

Spatial Join & Gap Analysis

Restaurants are spatially joined to SA2 polygons to count restaurants per suburb. An opportunity (gap) score is computed for each SA2 area.

Opportunity Score Formula

opportunity_score = (
  (india_pct / max_india_pct) * 0.4 +
  (median_income / max_income) * 0.2 +
  (1 - supply_per_10k / max_supply) * 0.4
)

Where supply_per_10k = number of Indian restaurants per 10,000 Indian-born residents in that SA2. A high score indicates high Indian population, decent income, but few restaurants — a market opportunity.

Gap Classification

Underserved: fewer than 4 restaurants per 10,000 Indian residents
Balanced: 4–8 restaurants per 10,000 Indian residents
Well-served: more than 8 restaurants per 10,000 Indian residents

Demand Score (per restaurant)

demand_score = rating * log10(review_count + 1)

This normalises for both rating quality and volume. A 4.5-star restaurant with 500 reviews scores higher than a 5-star restaurant with 3 reviews.

Deduplication

Restaurants appearing in multiple sources are deduplicated by matching on name (case-insensitive) and coordinates (rounded to 3 decimal places, ~100m). When a duplicate is found, the Google Places version is preferred (it has ratings and reviews), followed by Foursquare, then CLUE.

Limitations

Census data is from 2021. Melbourne's Indian population has grown significantly since then, particularly in western growth corridors (Tarneit, Truganina, Wyndham Vale). Real 2026 figures are likely 20–40% higher in these areas.
Google Places returns max 20 results per query. Suburbs with more than 20 Indian restaurants (e.g. Dandenong, Melbourne CBD) are under-counted.
CLUE data covers CBD only. The City of Melbourne CLUE dataset only includes businesses within the municipality of Melbourne. Outer council areas are not covered.
Foursquare data has no ratings. Foursquare OSP does not include ratings or review counts, so these restaurants show 0 for both fields.
Cuisine classification is heuristic. Sub-cuisine (South Indian vs North Indian etc.) is inferred from restaurant names using keyword matching. This is imperfect — a restaurant called "Spice Village" is classified as General even if it serves South Indian food.
Some false positives may exist. Keyword-based filtering from Foursquare may include non-Indian restaurants that happen to have matching words in their name (e.g. "Spice" in a non-Indian context).

Tech Stack

Map: MapLibre GL JS — open-source map rendering with dark CartoDB base tiles
Data pipeline: Python (pandas, geopandas, pyarrow, requests, shapely)
Dashboard: Single HTML file with vanilla JavaScript — no build step, no framework dependencies
Hosting: Static files on GitHub Pages

Source Code

The data pipeline (prepare_data.py) and dashboard source are open source. The pipeline fetches live data from all sources and generates the data/dashboard_data.js file that powers the map.

Disclaimer: This is an independent data visualisation project for educational and analytical purposes. It is not affiliated with the Australian Bureau of Statistics, Google, Foursquare, or the City of Melbourne. Restaurant data may be incomplete, outdated, or contain errors. Do not use this data as the sole basis for business decisions.